# Copyright (c) OpenMMLab. All rights reserved. import copy from typing import Dict, List, Tuple from torch import Tensor from mmcv.cnn import Linear from mmengine.model import bias_init_with_prob, constant_init from mmengine.structures import InstanceData from mmengine.model import BaseModule from mmdet.registry import MODELS from mmdet.structures import SampleList from mmdet.structures.bbox import bbox_cxcywh_to_xyxy, bbox_xyxy_to_cxcywh, bbox_overlaps from mmdet.utils import InstanceList, OptInstanceList, reduce_mean from ..utils import multi_apply from ..layers import inverse_sigmoid from .detr_head import DETRHead from mmdet.registry import MODELS, TASK_UTILS from mmdet.utils import (ConfigType, InstanceList, OptInstanceList,OptConfigType, OptMultiConfig, reduce_mean) from mmcv.ops import nms, batched_nms import torch import torch.nn as nn import torch.nn.functional as F from torch.nn import Transformer from ..losses import QualityFocalLoss # def adjust_bbox_to_pixel(bboxes: Tensor): # # 向下取整得到目标的左上角坐标 # adjusted_bboxes = torch.floor(bboxes) # # 向上取整得到目标的右下角坐标 # adjusted_bboxes[:, 2:] = torch.ceil(bboxes[:, 2:]) # return adjusted_bboxes def adjust_bbox_to_pixel(bboxes: Tensor): # 四舍五入取整坐标 adjusted_bboxes = torch.round(bboxes) return adjusted_bboxes @MODELS.register_module() class EvloveDetHead(BaseModule): r"""Head of the DINO: DETR with Improved DeNoising Anchor Boxes for End-to-End Object Detection Code is modified from the `official github repo `_. More details can be found in the `paper `_ . """ def __init__(self, num_classes: int, embed_dims: int = 256, num_reg_fcs: int = 2, center_feat_indice: int=1, sync_cls_avg_factor: bool = False, use_nms: bool = False, iou_threshold: float = 0.01, score_threshold: float = 0.0, class_wise_nms: bool = True, test_nms: OptConfigType = dict(type='nms', iou_threshold=0.01, ), neg_cls: bool = True, loss_cls: ConfigType = dict( type='CrossEntropyLoss', bg_cls_weight=0.1, use_sigmoid=False, loss_weight=1.0, class_weight=1.0), loss_center_cls: ConfigType = dict( type='CrossEntropyLoss', bg_cls_weight=0.1, use_sigmoid=False, loss_weight=1.0, class_weight=1.0), loss_bbox: ConfigType = dict(type='L1Loss', loss_weight=5.0), loss_iou: ConfigType = dict(type='GIoULoss', loss_weight=2.0), train_cfg: ConfigType = dict( assigner=dict( type='HungarianAssigner', match_costs=[ dict(type='ClassificationCost', weight=1.), dict(type='BBoxL1Cost', weight=5.0, box_format='xywh'), dict(type='IoUCost', iou_mode='giou', weight=2.0) ])), bbox_assigner: ConfigType = dict(type='MaxIoUAssigner', pos_iou_thr=0.01, neg_iou_thr=0.01, min_pos_iou=0.01, match_low_quality=False, ignore_iof_thr=-1), dn_assigner: ConfigType = dict(type='MaxIoUAssigner', pos_iou_thr=0.9, neg_iou_thr=0.9, min_pos_iou=0.9, match_low_quality=False, ignore_iof_thr=-1), test_cfg: ConfigType = dict(max_per_img=100), init_cfg: OptMultiConfig = None, share_pred_layer: bool = False, num_pred_layer: int = 6, as_two_stage: bool = False, pre_bboxes_round: bool = True, decoupe_dn: bool = False, dn_only_pos: bool = False, dn_loss_weight: List[float] = [1, 1, 1], # [1,1,1] center_neg_hard_num: int = 300, center_neg_rand_num: int = 300, loss_center_th: float = 0.2, loss_iou_th: float = 0.3, center_ds_ratio: int = 1 ) -> None: self.share_pred_layer = share_pred_layer self.num_pred_layer = num_pred_layer self.as_two_stage = as_two_stage self.pre_bboxes_round = pre_bboxes_round assert not (decoupe_dn and dn_only_pos), "Both decoupe_dn and dn_only_pos cannot be True at the same time." self.decoupe_dn = decoupe_dn self.dn_only_pos = dn_only_pos self.dn_loss_weight = dn_loss_weight self.iou_threshold = iou_threshold self.loss_center_th = loss_center_th self.loss_iou_th = loss_iou_th self.center_feat_indice = center_feat_indice self.center_ds_ratio = center_ds_ratio super().__init__(init_cfg=init_cfg) self.bg_cls_weight = 0 self.sync_cls_avg_factor = sync_cls_avg_factor class_weight = loss_cls.get('class_weight', None) if class_weight is not None and (self.__class__ is DETRHead): assert isinstance(class_weight, float), 'Expected ' \ 'class_weight to have type float. Found ' \ f'{type(class_weight)}.' # NOTE following the official DETR repo, bg_cls_weight means # relative classification weight of the no-object class. bg_cls_weight = loss_cls.get('bg_cls_weight', class_weight) assert isinstance(bg_cls_weight, float), 'Expected ' \ 'bg_cls_weight to have type float. Found ' \ f'{type(bg_cls_weight)}.' class_weight = torch.ones(num_classes + 1) * class_weight # set background class as the last indice class_weight[num_classes] = bg_cls_weight loss_cls.update({'class_weight': class_weight}) if 'bg_cls_weight' in loss_cls: loss_cls.pop('bg_cls_weight') self.bg_cls_weight = bg_cls_weight if train_cfg: assert 'assigner' in train_cfg, 'assigner should be provided ' \ 'when train_cfg is set.' assigner = train_cfg['assigner'] self.assigner = TASK_UTILS.build(assigner) if train_cfg.get('sampler', None) is not None: raise RuntimeError('DETR do not build sampler.') self.bbox_assigner = TASK_UTILS.build(bbox_assigner) self.dn_assigner = TASK_UTILS.build(dn_assigner) self.num_classes = num_classes self.embed_dims = embed_dims self.num_reg_fcs = num_reg_fcs self.train_cfg = train_cfg self.test_cfg = test_cfg self.loss_cls = MODELS.build(loss_cls) self.loss_center_cls = MODELS.build(loss_center_cls) self.loss_bbox = MODELS.build(loss_bbox) self.loss_iou = MODELS.build(loss_iou) self.use_nms = use_nms self.class_wise_nms = class_wise_nms self.score_threshold = score_threshold self.test_nms = test_nms self.neg_cls = neg_cls if self.loss_cls.use_sigmoid: self.cls_out_channels = num_classes else: self.cls_out_channels = num_classes + 1 self.center_neg_hard_num = center_neg_hard_num self.center_neg_rand_num = center_neg_rand_num self._init_layers() def _init_layers(self) -> None: """Initialize classification branch and regression branch of head.""" # fc_cls = Linear(self.embed_dims, self.cls_out_channels) fc_cls = [] # for _ in range(self.num_reg_fcs): # fc_cls.append(Linear(self.embed_dims, self.embed_dims)) # fc_cls.append(nn.ReLU()) fc_cls.append(Linear(self.embed_dims, self.cls_out_channels)) # fc_cls.append(Linear(self.embed_dims, 1)) fc_cls = nn.Sequential(*fc_cls) self.cls_branches = nn.ModuleList([fc_cls]) reg_branch = [] for _ in range(self.num_reg_fcs): reg_branch.append(Linear(self.embed_dims, self.embed_dims)) reg_branch.append(nn.ReLU()) reg_branch.append(Linear(self.embed_dims, 4)) reg_branch = nn.Sequential(*reg_branch) if self.share_pred_layer: self.reg_branches = nn.ModuleList( [reg_branch for _ in range(self.num_pred_layer)]) else: self.reg_branches = nn.ModuleList([ copy.deepcopy(reg_branch) for _ in range(self.num_pred_layer) ]) center_cls = [] # for _ in range(self.num_reg_fcs): # center_cls.append(Linear(self.embed_dims, self.embed_dims)) # center_cls.append(nn.ReLU()) center_cls.append(Linear(self.embed_dims, 1)) center_cls = nn.Sequential(*center_cls) self.center_cls = center_cls self.cls_embedding = nn.Embedding(1, self.embed_dims) # self.cls_embedding = nn.Embedding(self.cls_out_channels, self.embed_dims) def init_weights(self) -> None: """Initialize weights of the Deformable DETR head.""" if self.loss_cls.use_sigmoid: bias_init = bias_init_with_prob(0.01) for m in self.cls_branches: nn.init.constant_(m.bias, bias_init) nn.init.constant_(self.center_cls.bias, bias_init) for m in self.reg_branches: constant_init(m[-1], 0, bias=0) nn.init.constant_(self.reg_branches[0][-1].bias.data[2:], -2.0) if self.as_two_stage: for m in self.reg_branches: nn.init.constant_(m[-1].bias.data[2:], 0.0) nn.init.xavier_uniform_(self.cls_embedding) def _get_targets_single(self, cls_score: Tensor, bbox_pred: Tensor, gt_instances: InstanceData, img_meta: dict, with_neg_cls:bool=True, assigner_type:str = None) -> tuple: """Compute regression and classification targets for one image. Outputs from a single decoder layer of a single feature level are used. Args: cls_score (Tensor): Box score logits from a single decoder layer for one image. Shape [num_queries, cls_out_channels]. bbox_pred (Tensor): Sigmoid outputs from a single decoder layer for one image, with normalized coordinate (cx, cy, w, h) and shape [num_queries, 4]. gt_instances (:obj:`InstanceData`): Ground truth of instance annotations. It should includes ``bboxes`` and ``labels`` attributes. img_meta (dict): Meta information for one image. Returns: tuple[Tensor]: a tuple containing the following for one image. - labels (Tensor): Labels of each image. - label_weights (Tensor]): Label weights of each image. - bbox_targets (Tensor): BBox targets of each image. - bbox_weights (Tensor): BBox weights of each image. - pos_inds (Tensor): Sampled positive indices for each image. - neg_inds (Tensor): Sampled negative indices for each image. """ img_h, img_w = img_meta['img_shape'] factor = bbox_pred.new_tensor([img_w, img_h, img_w, img_h]).unsqueeze(0) num_bboxes = bbox_pred.size(0) # convert bbox_pred from xywh, normalized to xyxy, unnormalized bbox_pred = bbox_cxcywh_to_xyxy(bbox_pred) bbox_pred = bbox_pred * factor pred_instances = InstanceData(scores=cls_score, bboxes=bbox_pred,priors=bbox_pred) # assigner and sampler if assigner_type == 'dn': assign_result = self.dn_assigner.assign( pred_instances=pred_instances, gt_instances=gt_instances, img_meta=img_meta) else: assign_result = self.assigner.assign( pred_instances=pred_instances, gt_instances=gt_instances, img_meta=img_meta) gt_bboxes = gt_instances.bboxes gt_labels = gt_instances.labels pos_inds = torch.nonzero( assign_result.gt_inds > 0, as_tuple=False).squeeze(-1).unique() neg_inds = torch.nonzero( assign_result.gt_inds == 0, as_tuple=False).squeeze(-1).unique() pos_assigned_gt_inds = assign_result.gt_inds[pos_inds] - 1 pos_gt_bboxes = gt_bboxes[pos_assigned_gt_inds.long(), :] # label targets labels = gt_bboxes.new_full((num_bboxes,), self.num_classes, dtype=torch.long) labels[pos_inds] = gt_labels[pos_assigned_gt_inds] label_weights = gt_bboxes.new_zeros(num_bboxes) label_weights[pos_inds] = 1 label_weights[neg_inds] = 1 if not with_neg_cls: label_weights[neg_inds] = 0 # bbox targets bbox_targets = torch.zeros_like(bbox_pred) bbox_weights = torch.zeros_like(bbox_pred) bbox_weights[pos_inds] = 1.0 # DETR regress the relative position of boxes (cxcywh) in the image. # Thus the learning target should be normalized by the image size, also # the box format should be converted from defaultly x1y1x2y2 to cxcywh. pos_gt_bboxes_normalized = pos_gt_bboxes / factor pos_gt_bboxes_targets = bbox_xyxy_to_cxcywh(pos_gt_bboxes_normalized) bbox_targets[pos_inds] = pos_gt_bboxes_targets return (labels, label_weights, bbox_targets, bbox_weights, pos_inds, neg_inds) # def _get_targets_single_center(self, center_score: Tensor, center: Tensor, # spatial_shapes: Tensor, # gt_instances: InstanceData, # img_meta: dict) -> tuple: # """Compute regression and classification targets for one image. # # Outputs from a single decoder layer of a single feature level are used. # # Args: # cls_score (Tensor): Box score logits from a single decoder layer # for one image. Shape [num_queries, cls_out_channels]. # bbox_pred (Tensor): Sigmoid outputs from a single decoder layer # for one image, with normalized coordinate (cx, cy, w, h) and # shape [num_queries, 4]. # gt_instances (:obj:`InstanceData`): Ground truth of instance # annotations. It should includes ``bboxes`` and ``labels`` # attributes. # img_meta (dict): Meta information for one image. # # Returns: # tuple[Tensor]: a tuple containing the following for one image. # # - labels (Tensor): Labels of each image. # - label_weights (Tensor]): Label weights of each image. # - bbox_targets (Tensor): BBox targets of each image. # - bbox_weights (Tensor): BBox weights of each image. # - pos_inds (Tensor): Sampled positive indices for each image. # - neg_inds (Tensor): Sampled negative indices for each image. # """ # img_h, img_w = img_meta['img_shape'] # feat_w = int(spatial_shapes[self.center_feat_indice][0]) # feat_h = int(spatial_shapes[self.center_feat_indice][1]) # factor = center.new_tensor([feat_w, feat_h]).unsqueeze(0) # # factor = center.new_tensor([img_w, img_h,]).unsqueeze(0) # num_center = center.size(0) # # convert bbox_pred from xywh, normalized to xyxy, unnormalized # # bbox_pred = center # center = center * factor # gt_bboxes = gt_instances.bboxes # gt_cxcy = bbox_xyxy_to_cxcywh(gt_bboxes)[:, :2] # gt_cxcy[:, 0] = gt_cxcy[:, 0] * feat_w / img_w # gt_cxcy[:, 1] = gt_cxcy[:, 1] * feat_h / img_h # gt_cxcy_int = torch.floor(gt_cxcy+0.5)-0.5 # grid_y, grid_x = torch.meshgrid( # torch.linspace(-1, 1, 3, dtype=gt_cxcy.dtype, device=gt_cxcy.device), # torch.linspace(-1, 1, 3, dtype=gt_cxcy.dtype, device=gt_cxcy.device)) # grid = torch.cat([grid_x.unsqueeze(-1), grid_y.unsqueeze(-1)], -1) # grid = grid.view(-1, 2).unsqueeze(0).repeat(gt_cxcy.size(0), 1, 1) # candidate_point = gt_cxcy_int.unsqueeze(1).repeat(1, grid.size(1), 1)+grid # gt_cxcy_expand = gt_cxcy.unsqueeze(1).repeat(1, grid.size(1), 1) # dis = torch.sum((candidate_point-gt_cxcy_expand)**2, dim=2) # dis_min_indices = dis.argmin(dim=1) # dis_min_indices += torch.arange(len(dis_min_indices), device=dis_min_indices.device) * grid.size(1) # candidate_point = candidate_point.view(-1, 2) # target_point = candidate_point[dis_min_indices, :] # target_point[target_point<0.5]=0.5 # pos_index = torch.floor(target_point[:, 0]) + torch.floor(target_point[:, 1]) * feat_w # pos_index = pos_index.to(torch.long) # mask = (candidate_point > 0).all(dim=1) & (candidate_point[:, 0] <= feat_w) & (candidate_point[:, 1] <= feat_h) # candidate_point = candidate_point[mask] # candidate_index = torch.floor(candidate_point[:, 0]) + torch.floor(candidate_point[:, 1]) * feat_w # candidate_index = candidate_index.to(torch.long) # candidate_index = torch.unique(candidate_index) # _, indices = torch.sort(center_score,dim=0, descending=True) # sorted_indices = indices.squeeze() # mask = torch.isin(sorted_indices, candidate_index) # remaining_index = sorted_indices[~mask] # neg_index1 = remaining_index[:self.center_neg_hard_num] # neg_index2 = torch.randperm(remaining_index[self.center_neg_hard_num:].size(0), device=center.device)[:self.center_neg_rand_num ] # # neg_index = torch.cat([neg_index1, neg_index2], dim=0) # neg_index = neg_index2 # new_index = torch.cat([pos_index, neg_index], dim=0) # # new_center = center[new_index] # new_center_score = center_score[new_index] # pos_inds = torch.arange(0, pos_index.size(0)) # neg_inds = torch.arange(pos_index.size(0), new_index.size(0)) # labels = gt_bboxes.new_full((new_index.size(0),), 1, dtype=torch.long) # labels[:pos_index.size(0),] = 0 # label_weights = gt_bboxes.new_ones(new_index.size(0)) # # return (labels, label_weights, pos_inds, neg_inds, new_center_score) def _get_targets_single_center(self, center_score: Tensor, center: Tensor, spatial_shapes: Tensor, gt_instances: InstanceData, img_meta: dict) -> tuple: """Compute regression and classification targets for one image. Outputs from a single decoder layer of a single feature level are used. Args: cls_score (Tensor): Box score logits from a single decoder layer for one image. Shape [num_queries, cls_out_channels]. bbox_pred (Tensor): Sigmoid outputs from a single decoder layer for one image, with normalized coordinate (cx, cy, w, h) and shape [num_queries, 4]. gt_instances (:obj:`InstanceData`): Ground truth of instance annotations. It should includes ``bboxes`` and ``labels`` attributes. img_meta (dict): Meta information for one image. Returns: tuple[Tensor]: a tuple containing the following for one image. - labels (Tensor): Labels of each image. - label_weights (Tensor]): Label weights of each image. - bbox_targets (Tensor): BBox targets of each image. - bbox_weights (Tensor): BBox weights of each image. - pos_inds (Tensor): Sampled positive indices for each image. - neg_inds (Tensor): Sampled negative indices for each image. """ img_h, img_w = img_meta['img_shape'] feat_h = int(spatial_shapes[self.center_feat_indice][0]/self.center_ds_ratio) feat_w = int(spatial_shapes[self.center_feat_indice][1]/self.center_ds_ratio) factor = center.new_tensor([feat_w, feat_h]).unsqueeze(0) # factor = center.new_tensor([img_w, img_h,]).unsqueeze(0) num_center = center.size(0) # convert bbox_pred from xywh, normalized to xyxy, unnormalized # bbox_pred = center center = center * factor gt_bboxes = gt_instances.bboxes gt_cxcy = bbox_xyxy_to_cxcywh(gt_bboxes)[:, :2] gt_cxcy[:, 0] = gt_cxcy[:, 0] * feat_w / img_w gt_cxcy[:, 1] = gt_cxcy[:, 1] * feat_h / img_h gt_cxcy= gt_cxcy.long() gt_bboxes[:, 2:] -= 0.1 gt_bboxes_x = gt_bboxes[:, 0::2] gt_bboxes_y = gt_bboxes[:, 1::2] gt_bboxes_x = torch.floor(gt_bboxes_x * feat_w / img_w) gt_bboxes_y = torch.floor(gt_bboxes_y * feat_h / img_h) gt_bboxes_x = gt_bboxes_x.long() gt_bboxes_y = gt_bboxes_y.long() heat_map = gt_bboxes.new_full((feat_h, feat_w), 0, dtype=torch.long) for t_i in range(gt_bboxes.size(0)): grid_y, grid_x = torch.meshgrid( torch.linspace(gt_bboxes_y[t_i, 0], gt_bboxes_y[t_i, 1], gt_bboxes_y[t_i, 1]+1-gt_bboxes_y[t_i, 0], dtype=torch.long, device=gt_cxcy.device), torch.linspace(gt_bboxes_x[t_i, 0], gt_bboxes_x[t_i, 1], gt_bboxes_x[t_i, 1]+1-gt_bboxes_x[t_i, 0], dtype=torch.long, device=gt_cxcy.device)) grid = torch.cat([grid_y.unsqueeze(-1), grid_x.unsqueeze(-1)], -1) grid = grid.view(-1, 2) a = gt_bboxes.new_full((grid.size(0),), -1, dtype=torch.long) heat_map.index_put_((grid[:,0],grid[:,1]), a) a = gt_bboxes.new_full((gt_cxcy.size(0),), 1, dtype=torch.long) heat_map = heat_map.index_put_((gt_cxcy[:,1], gt_cxcy[:,0]), a) heat_map = heat_map.view(-1) mask = heat_map != -1 new_center_score = center_score[mask] heat_map = heat_map[mask] pos_inds = torch.where(heat_map == 1)[0] neg_inds = torch.where(heat_map == 0)[0] labels = gt_bboxes.new_full((heat_map.size(0),), 1, dtype=torch.long) labels[pos_inds] = 0 label_weights = gt_bboxes.new_ones(heat_map.size(0)) return (labels, label_weights, pos_inds, neg_inds, new_center_score) def _get_targets_single_bbox(self,cls_score: Tensor, bbox_pred: Tensor, gt_instances: InstanceData, img_meta: dict) -> tuple: """Compute regression and classification targets for one image. Outputs from a single decoder layer of a single feature level are used. Args: cls_score (Tensor): Box score logits from a single decoder layer for one image. Shape [num_queries, cls_out_channels]. bbox_pred (Tensor): Sigmoid outputs from a single decoder layer for one image, with normalized coordinate (cx, cy, w, h) and shape [num_queries, 4]. gt_instances (:obj:`InstanceData`): Ground truth of instance annotations. It should includes ``bboxes`` and ``labels`` attributes. img_meta (dict): Meta information for one image. Returns: tuple[Tensor]: a tuple containing the following for one image. - labels (Tensor): Labels of each image. - label_weights (Tensor]): Label weights of each image. - bbox_targets (Tensor): BBox targets of each image. - bbox_weights (Tensor): BBox weights of each image. - pos_inds (Tensor): Sampled positive indices for each image. - neg_inds (Tensor): Sampled negative indices for each image. """ img_h, img_w = img_meta['img_shape'] factor = bbox_pred.new_tensor([img_w, img_h, img_w, img_h]).unsqueeze(0) num_bboxes = bbox_pred.size(0) # convert bbox_pred from xywh, normalized to xyxy, unnormalized bbox_pred = bbox_cxcywh_to_xyxy(bbox_pred) bbox_pred = bbox_pred * factor pred_instances = InstanceData(scores=cls_score, bboxes=bbox_pred,priors=bbox_pred) # assigner and sampler assign_result = self.bbox_assigner.assign( pred_instances=pred_instances, gt_instances=gt_instances, img_meta=img_meta) gt_bboxes = gt_instances.bboxes gt_labels = gt_instances.labels pos_inds = torch.nonzero( assign_result.gt_inds > 0, as_tuple=False).squeeze(-1).unique() neg_inds = torch.nonzero( assign_result.gt_inds == 0, as_tuple=False).squeeze(-1).unique() pos_assigned_gt_inds = assign_result.gt_inds[pos_inds] - 1 pos_gt_bboxes = gt_bboxes[pos_assigned_gt_inds.long(), :] # bbox targets bbox_targets = torch.zeros_like(bbox_pred) bbox_weights = torch.zeros_like(bbox_pred) bbox_weights[pos_inds] = 1.0 # DETR regress the relative position of boxes (cxcywh) in the image. # Thus the learning target should be normalized by the image size, also # the box format should be converted from defaultly x1y1x2y2 to cxcywh. pos_gt_bboxes_normalized = pos_gt_bboxes / factor pos_gt_bboxes_targets = bbox_xyxy_to_cxcywh(pos_gt_bboxes_normalized) bbox_targets[pos_inds] = pos_gt_bboxes_targets return (bbox_targets, bbox_weights, pos_inds, neg_inds) def loss_and_predict( self, hidden_states: Tuple[Tensor], batch_data_samples: SampleList) -> Tuple[dict, InstanceList]: """Perform forward propagation of the head, then calculate loss and predictions from the features and data samples. Over-write because img_metas are needed as inputs for bbox_head. Args: hidden_states (tuple[Tensor]): Feature from the transformer decoder, has shape (num_decoder_layers, bs, num_queries, dim). batch_data_samples (list[:obj:`DetDataSample`]): Each item contains the meta information of each image and corresponding annotations. Returns: tuple: the return value is a tuple contains: - losses: (dict[str, Tensor]): A dictionary of loss components. - predictions (list[:obj:`InstanceData`]): Detection results of each image after the post process. """ batch_gt_instances = [] batch_img_metas = [] for data_sample in batch_data_samples: batch_img_metas.append(data_sample.metainfo) batch_gt_instances.append(data_sample.gt_instances) outs = self(hidden_states) loss_inputs = outs + (batch_gt_instances, batch_img_metas) losses = self.loss_by_feat(*loss_inputs) predictions = self.predict_by_feat( *outs, batch_img_metas=batch_img_metas) return losses, predictions def forward(self, hidden_states: Tensor, references: List[Tensor], cls_feats: Tensor,) -> Tuple[Tensor]: """Forward function. Args: hidden_states (Tensor): Hidden states output from each decoder layer, has shape (num_decoder_layers, bs, num_queries, dim). references (list[Tensor]): List of the reference from the decoder. The first reference is the `init_reference` (initial) and the other num_decoder_layers(6) references are `inter_references` (intermediate). The `init_reference` has shape (bs, num_queries, 4) when `as_two_stage` of the detector is `True`, otherwise (bs, num_queries, 2). Each `inter_reference` has shape (bs, num_queries, 4) when `with_box_refine` of the detector is `True`, otherwise (bs, num_queries, 2). The coordinates are arranged as (cx, cy) when the last dimension is 2, and (cx, cy, w, h) when it is 4. Returns: tuple[Tensor]: results of head containing the following tensor. - all_layers_outputs_classes (Tensor): Outputs from the classification head, has shape (num_decoder_layers, bs, num_queries, cls_out_channels). - all_layers_outputs_coords (Tensor): Sigmoid outputs from the regression head with normalized coordinate format (cx, cy, w, h), has shape (num_decoder_layers, bs, num_queries, 4) with the last dimension arranged as (cx, cy, w, h). """ all_layers_outputs_coords = [] for layer_id in range(hidden_states.shape[0]): reference = inverse_sigmoid(references[layer_id]) # NOTE The last reference will not be used. hidden_state = hidden_states[layer_id] tmp_reg_preds = self.reg_branches[layer_id](hidden_state) if reference.shape[-1] == 4: # When `layer` is 0 and `as_two_stage` of the detector # is `True`, or when `layer` is greater than 0 and # `with_box_refine` of the detector is `True`. tmp_reg_preds += reference else: # When `layer` is 0 and `as_two_stage` of the detector # is `False`, or when `layer` is greater than 0 and # `with_box_refine` of the detector is `False`. assert reference.shape[-1] == 2 tmp_reg_preds[..., :2] += reference outputs_coord = tmp_reg_preds.sigmoid() all_layers_outputs_coords.append(outputs_coord) # cls_embedding = self.cls_embedding.weight[:, None, :] # cls_embedding = cls_embedding.transpose(0, 1).unsqueeze(0).repeat(cls_feats.size(0),cls_feats.size(1),1,1) # cls_feats = cls_feats.unsqueeze(2).repeat(1, 1, self.cls_out_channels, 1) # cls_feats = cls_feats+cls_embedding # cls_feats = cls_feats.reshape(cls_feats.size(0), -1, cls_feats.size(3)) # outputs_classes = self.cls_branches[-1](cls_feats) # outputs_classes = outputs_classes.squeeze(dim=-1).transpose(1, 2).view(cls_feats.size(0),-1,self.cls_out_channels) # cls_feats = cls_feats *self.cls_embedding.weight[:, None, :] outputs_classes = self.cls_branches[-1](cls_feats) all_layers_outputs_classes = outputs_classes.unsqueeze(0).repeat(hidden_states.shape[0],1,1,1) all_layers_outputs_coords = torch.stack(all_layers_outputs_coords) inputs_coords = references[0] return outputs_classes, inputs_coords, all_layers_outputs_classes, all_layers_outputs_coords def loss(self, hidden_states: Tensor, references: List[Tensor], centers: Tensor, center_scores: Tensor, topk_centers_scores: Tensor, cls_feats: Tensor, batch_data_samples: SampleList, dn_meta: Dict[str, int], spatial_shapes: Tensor) -> dict: """Perform forward propagation and loss calculation of the detection head on the queries of the upstream network. Args: hidden_states (Tensor): Hidden states output from each decoder layer, has shape (num_decoder_layers, bs, num_queries_total, dim), where `num_queries_total` is the sum of `num_denoising_queries` and `num_matching_queries` when `self.training` is `True`, else `num_matching_queries`. references (list[Tensor]): List of the reference from the decoder. The first reference is the `init_reference` (initial) and the other num_decoder_layers(6) references are `inter_references` (intermediate). The `init_reference` has shape (bs, num_queries_total, 4) and each `inter_reference` has shape (bs, num_queries, 4) with the last dimension arranged as (cx, cy, w, h). enc_outputs_class (Tensor): The score of each point on encode feature map, has shape (bs, num_feat_points, cls_out_channels). enc_outputs_coord (Tensor): The proposal generate from the encode feature map, has shape (bs, num_feat_points, 4) with the last dimension arranged as (cx, cy, w, h). batch_data_samples (list[:obj:`DetDataSample`]): The Data Samples. It usually includes information such as `gt_instance`, `gt_panoptic_seg` and `gt_sem_seg`. dn_meta (Dict[str, int]): The dictionary saves information about group collation, including 'num_denoising_queries' and 'num_denoising_groups'. It will be used for split outputs of denoising and matching parts and loss calculation. Returns: dict: A dictionary of loss components. """ batch_gt_instances = [] batch_img_metas = [] for data_sample in batch_data_samples: batch_img_metas.append(data_sample.metainfo) batch_gt_instances.append(data_sample.gt_instances) outs = self(hidden_states, references, cls_feats) loss_inputs = outs + (center_scores, centers, topk_centers_scores, batch_gt_instances, batch_img_metas, dn_meta, spatial_shapes) losses = self.loss_by_feat(*loss_inputs) return losses def loss_by_feat( self, cls_scores: Tensor, inputs_coords: Tensor, all_layers_cls_scores: Tensor, all_layers_bbox_preds: Tensor, center_scores: Tensor, centers: Tensor, topk_centers_scores: Tensor, batch_gt_instances: InstanceList, batch_img_metas: List[dict], dn_meta: Dict[str, int], spatial_shapes: Tensor, batch_gt_instances_ignore: OptInstanceList = None ) -> Dict[str, Tensor]: """Loss function. Args: all_layers_cls_scores (Tensor): Classification scores of all decoder layers, has shape (num_decoder_layers, bs, num_queries_total, cls_out_channels), where `num_queries_total` is the sum of `num_denoising_queries` and `num_matching_queries`. all_layers_bbox_preds (Tensor): Regression outputs of all decoder layers. Each is a 4D-tensor with normalized coordinate format (cx, cy, w, h) and has shape (num_decoder_layers, bs, num_queries_total, 4). enc_cls_scores (Tensor): The score of each point on encode feature map, has shape (bs, num_feat_points, cls_out_channels). enc_bbox_preds (Tensor): The proposal generate from the encode feature map, has shape (bs, num_feat_points, 4) with the last dimension arranged as (cx, cy, w, h). batch_gt_instances (list[:obj:`InstanceData`]): Batch of gt_instance. It usually includes ``bboxes`` and ``labels`` attributes. batch_img_metas (list[dict]): Meta information of each image, e.g., image size, scaling factor, etc. dn_meta (Dict[str, int]): The dictionary saves information about group collation, including 'num_denoising_queries' and 'num_denoising_groups'. It will be used for split outputs of denoising and matching parts and loss calculation. batch_gt_instances_ignore (list[:obj:`InstanceData`], optional): Batch of gt_instances_ignore. It includes ``bboxes`` attribute data that is ignored during training and testing. Defaults to None. Returns: dict[str, Tensor]: A dictionary of loss components. """ # extract denoising and matching part of outputs weight_bbox = 0 weight_cls = 0 (all_layers_matching_cls_scores, all_layers_matching_bbox_preds, all_layers_denoising_cls_scores, all_layers_denoising_bbox_preds) = \ self.split_outputs(all_layers_cls_scores, all_layers_bbox_preds, dn_meta) (matching_cls_scores, all_layers_matching_bbox_preds, denoising_cls_scores, all_layers_denoising_bbox_preds) = \ self.split_outputsv1(cls_scores, all_layers_bbox_preds, dn_meta) loss_dict = dict() center_loss_cls = self.loss_center(center_scores, centers, spatial_shapes, batch_gt_instances=batch_gt_instances, batch_img_metas=batch_img_metas) loss_dict['center_loss_cls'] = center_loss_cls if center_loss_cls <= self.loss_center_th: weight_bbox = 1 # num_imgs = centers.size(0) # bbox_preds_list = [inputs_coords[i][dn_meta['num_denoising_queries']:] for i in range(num_imgs)] # cls_scores_list = [all_layers_matching_cls_scores[-1][i] for i in range(num_imgs)] # reg_targets = self.get_targets_bbox(cls_scores_list, bbox_preds_list, batch_gt_instances, batch_img_metas) # losses_bbox, losses_iou = multi_apply( # self.loss_bbox_by_feat_single, # all_layers_matching_bbox_preds, # reg_targets=reg_targets, # batch_gt_instances=batch_gt_instances, # batch_img_metas=batch_img_metas) # # loss from other decoder layers # num_dec_layer = 1 # for loss_bbox_i, loss_iou_i in zip(losses_bbox, losses_iou): # loss_dict[f'd{num_dec_layer}.loss_bbox'] = loss_bbox_i*weight_bbox # loss_dict[f'd{num_dec_layer}.loss_iou'] = loss_iou_i*weight_bbox # num_dec_layer += 1 reg_targets = self.get_dn_targets(batch_gt_instances, batch_img_metas, dn_meta) dn_losses_bbox, dn_losses_iou = multi_apply( self._loss_dn_single, all_layers_denoising_bbox_preds, reg_targets=reg_targets, batch_gt_instances=batch_gt_instances, batch_img_metas=batch_img_metas, dn_meta=dn_meta) for num_dec_layer, (loss_bbox_i, loss_iou_i) in \ enumerate(zip(dn_losses_bbox, dn_losses_iou)): loss_dict[f'd{num_dec_layer+1}.dn_loss_bbox'] = loss_bbox_i*weight_bbox loss_dict[f'd{num_dec_layer+1}.dn_loss_iou'] = loss_iou_i*weight_bbox if weight_bbox == 1 and loss_iou_i <= self.loss_iou_th: weight_cls = 1 # loss_cls = self.cls_loss(cls_scores, all_layers_bbox_preds[-1], batch_gt_instances=batch_gt_instances, # batch_img_metas=batch_img_metas) loss_cls = self.cls_loss(all_layers_matching_cls_scores[-1], all_layers_matching_bbox_preds[-1], batch_gt_instances=batch_gt_instances, batch_img_metas=batch_img_metas) # loss_cls = self.cls_loss_dn_match(matching_cls_scores, all_layers_matching_bbox_preds[-1], # denoising_cls_scores, all_layers_denoising_bbox_preds[-1], reg_targets, batch_gt_instances=batch_gt_instances, # batch_img_metas=batch_img_metas) loss_dict['match_loss_cls'] = loss_cls*weight_cls if self.cls_out_channels > 0: dn_loss_cls = self.cls_loss_dn(all_layers_denoising_cls_scores[-1], all_layers_denoising_bbox_preds[-1], reg_targets, batch_gt_instances=batch_gt_instances, batch_img_metas=batch_img_metas) loss_dict['dn_loss_cls'] = dn_loss_cls * weight_cls return loss_dict def loss_by_feat_single(self, cls_scores: Tensor, bbox_preds: Tensor, batch_gt_instances: InstanceList, batch_img_metas: List[dict]) -> Tuple[Tensor]: """Loss function for outputs from a single decoder layer of a single feature level. Args: cls_scores (Tensor): Box score logits from a single decoder layer for all images, has shape (bs, num_queries, cls_out_channels). bbox_preds (Tensor): Sigmoid outputs from a single decoder layer for all images, with normalized coordinate (cx, cy, w, h) and shape (bs, num_queries, 4). batch_gt_instances (list[:obj:`InstanceData`]): Batch of gt_instance. It usually includes ``bboxes`` and ``labels`` attributes. batch_img_metas (list[dict]): Meta information of each image, e.g., image size, scaling factor, etc. Returns: Tuple[Tensor]: A tuple including `loss_cls`, `loss_box` and `loss_iou`. """ num_imgs = cls_scores.size(0) cls_scores_list = [cls_scores[i] for i in range(num_imgs)] bbox_preds_list = [bbox_preds[i] for i in range(num_imgs)] cls_reg_targets = self.get_targets(cls_scores_list, bbox_preds_list, batch_gt_instances, batch_img_metas) (labels_list, label_weights_list, bbox_targets_list, bbox_weights_list, num_total_pos, num_total_neg) = cls_reg_targets labels = torch.cat(labels_list, 0) label_weights = torch.cat(label_weights_list, 0) bbox_targets = torch.cat(bbox_targets_list, 0) bbox_weights = torch.cat(bbox_weights_list, 0) # classification loss cls_scores = cls_scores.reshape(-1, self.cls_out_channels) # construct weighted avg_factor to match with the official DETR repo cls_avg_factor = num_total_pos * 1.0 + \ num_total_neg * self.bg_cls_weight if not self.neg_cls: cls_avg_factor = num_total_pos * 1.0 if self.sync_cls_avg_factor: cls_avg_factor = reduce_mean( cls_scores.new_tensor([cls_avg_factor])) cls_avg_factor = max(cls_avg_factor, 1) loss_cls = self.loss_cls( cls_scores, labels, label_weights, avg_factor=cls_avg_factor) # Compute the average number of gt boxes across all gpus, for # normalization purposes num_total_pos = loss_cls.new_tensor([num_total_pos]) num_total_pos = torch.clamp(reduce_mean(num_total_pos), min=1).item() # construct factors used for rescale bboxes factors = [] for img_meta, bbox_pred in zip(batch_img_metas, bbox_preds): img_h, img_w, = img_meta['img_shape'] factor = bbox_pred.new_tensor([img_w, img_h, img_w, img_h]).unsqueeze(0).repeat( bbox_pred.size(0), 1) factors.append(factor) factors = torch.cat(factors, 0) # DETR regress the relative position of boxes (cxcywh) in the image, # thus the learning target is normalized by the image size. So here # we need to re-scale them for calculating IoU loss bbox_preds = bbox_preds.reshape(-1, 4) bboxes = bbox_cxcywh_to_xyxy(bbox_preds) * factors bboxes_gt = bbox_cxcywh_to_xyxy(bbox_targets) * factors # regression IoU loss, defaultly GIoU loss loss_iou = self.loss_iou( bboxes, bboxes_gt, bbox_weights, avg_factor=num_total_pos) # regression L1 loss loss_bbox = self.loss_bbox( bbox_preds, bbox_targets, bbox_weights, avg_factor=num_total_pos) return loss_cls, loss_bbox, loss_iou def cls_loss(self, cls_scores: Tensor, bbox_preds: Tensor, batch_gt_instances: InstanceList, batch_img_metas: List[dict]) -> Tuple[Tensor]: """Loss function for outputs from a single decoder layer of a single feature level. Args: cls_scores (Tensor): Box score logits from a single decoder layer for all images, has shape (bs, num_queries, cls_out_channels). bbox_preds (Tensor): Sigmoid outputs from a single decoder layer for all images, with normalized coordinate (cx, cy, w, h) and shape (bs, num_queries, 4). batch_gt_instances (list[:obj:`InstanceData`]): Batch of gt_instance. It usually includes ``bboxes`` and ``labels`` attributes. batch_img_metas (list[dict]): Meta information of each image, e.g., image size, scaling factor, etc. Returns: Tuple[Tensor]: A tuple including `loss_cls`, `loss_box` and `loss_iou`. """ num_imgs = cls_scores.size(0) cls_scores_list = [cls_scores[i] for i in range(num_imgs)] bbox_preds_list = [bbox_preds[i] for i in range(num_imgs)] cls_reg_targets = self.get_targets(cls_scores_list, bbox_preds_list, batch_gt_instances, batch_img_metas) (labels_list, label_weights_list, bbox_targets_list, bbox_weights_list, num_total_pos, num_total_neg) = cls_reg_targets labels = torch.cat(labels_list, 0) label_weights = torch.cat(label_weights_list, 0) bbox_targets = torch.cat(bbox_targets_list, 0) bbox_weights = torch.cat(bbox_weights_list, 0) # classification loss cls_scores = cls_scores.reshape(-1, self.cls_out_channels) # construct weighted avg_factor to match with the official DETR repo cls_avg_factor = num_total_pos * 1.0 + \ num_total_neg * self.bg_cls_weight if self.sync_cls_avg_factor: cls_avg_factor = reduce_mean( cls_scores.new_tensor([cls_avg_factor])) cls_avg_factor = max(cls_avg_factor, 1) if isinstance(self.loss_cls, QualityFocalLoss): bg_class_ind = self.num_classes pos_inds = ((labels >= 0) & (labels < bg_class_ind)).nonzero().squeeze(1) scores = label_weights.new_zeros(labels.shape) pos_bbox_targets = bbox_targets[pos_inds] pos_decode_bbox_targets = bbox_cxcywh_to_xyxy(pos_bbox_targets) pos_bbox_pred = bbox_preds.reshape(-1, 4)[pos_inds] pos_decode_bbox_pred = bbox_cxcywh_to_xyxy(pos_bbox_pred) scores[pos_inds] = bbox_overlaps( pos_decode_bbox_pred.detach(), pos_decode_bbox_targets, is_aligned=True) loss_cls = self.loss_cls( cls_scores, (labels, scores), label_weights, avg_factor=cls_avg_factor) else: loss_cls = self.loss_cls( cls_scores, labels, label_weights, avg_factor=cls_avg_factor) # loss_cls = self.loss_cls( # cls_scores, labels, label_weights, avg_factor=cls_avg_factor) if num_total_pos == 0: loss_cls = loss_cls*0 return loss_cls def cls_loss_dn_match(self, matching_cls_scores: Tensor, matching_bbox_preds: Tensor, denoising_cls_scores: Tensor, denoising_bbox_preds: Tensor, dn_targets: Tuple[list, int], batch_gt_instances: InstanceList, batch_img_metas: List[dict]) -> Tuple[Tensor]: """Loss function for outputs from a single decoder layer of a single feature level. Args: cls_scores (Tensor): Box score logits from a single decoder layer for all images, has shape (bs, num_queries, cls_out_channels). bbox_preds (Tensor): Sigmoid outputs from a single decoder layer for all images, with normalized coordinate (cx, cy, w, h) and shape (bs, num_queries, 4). batch_gt_instances (list[:obj:`InstanceData`]): Batch of gt_instance. It usually includes ``bboxes`` and ``labels`` attributes. batch_img_metas (list[dict]): Meta information of each image, e.g., image size, scaling factor, etc. Returns: Tuple[Tensor]: A tuple including `loss_cls`, `loss_box` and `loss_iou`. """ cls_scores= torch.cat([matching_cls_scores,denoising_cls_scores],dim=1) bbox_preds = torch.cat([matching_bbox_preds, denoising_bbox_preds], dim=1) # cls_scores = matching_cls_scores # bbox_preds = matching_bbox_preds num_imgs = cls_scores.size(0) cls_scores_list = [cls_scores[i] for i in range(num_imgs)] bbox_preds_list = [bbox_preds[i] for i in range(num_imgs)] matching_cls_scores_list = [matching_cls_scores[i] for i in range(num_imgs)] matching_bbox_preds_list = [matching_bbox_preds[i] for i in range(num_imgs)] denoising_cls_scores_list = [denoising_cls_scores[i] for i in range(num_imgs)] denoising_bbox_preds_list = [denoising_bbox_preds[i] for i in range(num_imgs)] cls_reg_targets = self.get_targets(matching_cls_scores_list, matching_bbox_preds_list, batch_gt_instances, batch_img_metas) (labels_list, label_weights_list, bbox_targets_list, bbox_weights_list, num_total_pos, num_total_neg) = cls_reg_targets cls_reg_targets_dn = self.get_targets(denoising_cls_scores_list, denoising_bbox_preds_list, batch_gt_instances, batch_img_metas, with_neg_cls=False, assigner_type='dn') (labels_list_dn, label_weights_list_dn, bbox_targets_list_dn, bbox_weights_list_dn, num_total_pos_dn, num_total_neg_dn) = cls_reg_targets_dn # (labels_list_dn, label_weights_list_dn,bbox_targets_list_dn, bbox_weights_list_dn, num_total_pos_dn) = dn_targets for i in range(num_imgs): labels_list[i] = torch.cat([labels_list[i], labels_list_dn[i]], 0) label_weights_list[i] = torch.cat([label_weights_list[i], label_weights_list_dn[i]], 0) bbox_targets_list[i] = torch.cat([bbox_targets_list[i], bbox_targets_list_dn[i]], 0) bbox_weights_list[i] = torch.cat([bbox_weights_list[i], bbox_weights_list_dn[i]], 0) num_total_pos = num_total_pos+num_total_pos_dn labels = torch.cat(labels_list, 0) label_weights = torch.cat(label_weights_list, 0) bbox_targets = torch.cat(bbox_targets_list, 0) # bbox_weights = torch.cat(bbox_weights_list, 0) cls_scores = cls_scores.reshape(-1, self.cls_out_channels) labels = labels[label_weights != 0] cls_scores = cls_scores[label_weights != 0] bbox_targets = bbox_targets[label_weights != 0] label_weights = label_weights[label_weights!= 0] # construct weighted avg_factor to match with the official DETR repo cls_avg_factor = num_total_pos * 1.0 + \ num_total_neg * self.bg_cls_weight if self.sync_cls_avg_factor: cls_avg_factor = reduce_mean( cls_scores.new_tensor([cls_avg_factor])) cls_avg_factor = max(cls_avg_factor, 1) if isinstance(self.loss_cls, QualityFocalLoss): bg_class_ind = self.num_classes pos_inds = ((labels >= 0) & (labels < bg_class_ind)).nonzero().squeeze(1) scores = label_weights.new_zeros(labels.shape) pos_bbox_targets = bbox_targets[pos_inds] pos_decode_bbox_targets = bbox_cxcywh_to_xyxy(pos_bbox_targets) pos_bbox_pred = bbox_preds.reshape(-1, 4)[pos_inds] pos_decode_bbox_pred = bbox_cxcywh_to_xyxy(pos_bbox_pred) scores[pos_inds] = bbox_overlaps( pos_decode_bbox_pred.detach(), pos_decode_bbox_targets, is_aligned=True) loss_cls = self.loss_cls( cls_scores, (labels, scores), label_weights, avg_factor=cls_avg_factor) else: loss_cls = self.loss_cls( cls_scores, labels, label_weights, avg_factor=cls_avg_factor) # loss_cls = self.loss_cls( # cls_scores, labels, label_weights, avg_factor=cls_avg_factor) if num_total_pos == 0: loss_cls = loss_cls*0 return loss_cls def cls_loss_dn(self,denoising_cls_scores: Tensor, denoising_bbox_preds: Tensor, dn_targets: Tuple[list, int], batch_gt_instances: InstanceList, batch_img_metas: List[dict]) -> Tuple[Tensor]: """Loss function for outputs from a single decoder layer of a single feature level. Args: cls_scores (Tensor): Box score logits from a single decoder layer for all images, has shape (bs, num_queries, cls_out_channels). bbox_preds (Tensor): Sigmoid outputs from a single decoder layer for all images, with normalized coordinate (cx, cy, w, h) and shape (bs, num_queries, 4). batch_gt_instances (list[:obj:`InstanceData`]): Batch of gt_instance. It usually includes ``bboxes`` and ``labels`` attributes. batch_img_metas (list[dict]): Meta information of each image, e.g., image size, scaling factor, etc. Returns: Tuple[Tensor]: A tuple including `loss_cls`, `loss_box` and `loss_iou`. """ cls_scores = denoising_cls_scores bbox_preds = denoising_bbox_preds num_imgs = cls_scores.size(0) cls_scores_list = [cls_scores[i] for i in range(num_imgs)] bbox_preds_list = [bbox_preds[i] for i in range(num_imgs)] cls_reg_targets = self.get_targets(cls_scores_list, bbox_preds_list, batch_gt_instances, batch_img_metas, with_neg_cls=False, assigner_type='dn') (labels_list, label_weights_list, bbox_targets_list, bbox_weights_list, num_total_pos, num_total_neg) = cls_reg_targets # (labels_list, label_weights_list,bbox_targets_list, bbox_weights_list, num_total_pos) = dn_targets labels = torch.cat(labels_list, 0) label_weights = torch.cat(label_weights_list, 0) bbox_targets = torch.cat(bbox_targets_list, 0) bbox_weights = torch.cat(bbox_weights_list, 0) # classification loss cls_scores = cls_scores.reshape(-1, self.cls_out_channels) # construct weighted avg_factor to match with the official DETR repo cls_avg_factor = num_total_pos * 1.0 + \ num_total_neg * self.bg_cls_weight if self.sync_cls_avg_factor: cls_avg_factor = reduce_mean( cls_scores.new_tensor([cls_avg_factor])) cls_avg_factor = max(cls_avg_factor, 1) if isinstance(self.loss_cls, QualityFocalLoss): bg_class_ind = self.num_classes pos_inds = ((labels >= 0) & (labels < bg_class_ind)).nonzero().squeeze(1) scores = label_weights.new_zeros(labels.shape) pos_bbox_targets = bbox_targets[pos_inds] pos_decode_bbox_targets = bbox_cxcywh_to_xyxy(pos_bbox_targets) pos_bbox_pred = bbox_preds.reshape(-1, 4)[pos_inds] pos_decode_bbox_pred = bbox_cxcywh_to_xyxy(pos_bbox_pred) scores[pos_inds] = bbox_overlaps( pos_decode_bbox_pred.detach(), pos_decode_bbox_targets, is_aligned=True) loss_cls = self.loss_cls( cls_scores, (labels, scores), label_weights, avg_factor=cls_avg_factor) else: loss_cls = self.loss_cls( cls_scores, labels, label_weights, avg_factor=cls_avg_factor) # loss_cls = self.loss_cls( # cls_scores, labels, label_weights, avg_factor=cls_avg_factor) if num_total_pos == 0: loss_cls = loss_cls*0 return loss_cls def loss_center(self, center_scores: Tensor, centers: Tensor, spatial_shapes: Tensor, batch_gt_instances: InstanceList, batch_img_metas: List[dict]) -> Tuple[Tensor]: """Loss function for outputs from a single decoder layer of a single feature level. Args: cls_scores (Tensor): Box score logits from a single decoder layer for all images, has shape (bs, num_queries, cls_out_channels). bbox_preds (Tensor): Sigmoid outputs from a single decoder layer for all images, with normalized coordinate (cx, cy, w, h) and shape (bs, num_queries, 4). batch_gt_instances (list[:obj:`InstanceData`]): Batch of gt_instance. It usually includes ``bboxes`` and ``labels`` attributes. batch_img_metas (list[dict]): Meta information of each image, e.g., image size, scaling factor, etc. Returns: Tuple[Tensor]: A tuple including `loss_cls`, `loss_box` and `loss_iou`. """ num_imgs = center_scores.size(0) center_scores_list = [center_scores[i] for i in range(num_imgs)] centers_list = [centers[i] for i in range(num_imgs)] spatial_shapes_list = [spatial_shapes for i in range(num_imgs)] (labels_list, label_weights_list, pos_inds_list, neg_inds_list, new_center_scores_list) = multi_apply(self._get_targets_single_center, center_scores_list, centers_list, spatial_shapes_list, batch_gt_instances, batch_img_metas) num_total_pos = sum((inds.numel() for inds in pos_inds_list)) num_total_neg = sum((inds.numel() for inds in neg_inds_list)) labels = torch.cat(labels_list, 0) label_weights = torch.cat(label_weights_list, 0) new_center_scores = torch.cat(new_center_scores_list, 0) # construct weighted avg_factor to match with the official DETR repo cls_avg_factor = num_total_pos * 1.0 + \ num_total_neg * self.bg_cls_weight if self.sync_cls_avg_factor: cls_avg_factor = reduce_mean( new_center_scores.new_tensor([cls_avg_factor])) cls_avg_factor = max(cls_avg_factor, 1) loss_cls = self.loss_center_cls( new_center_scores, labels, label_weights, avg_factor=cls_avg_factor) return loss_cls def loss_bbox_by_feat_single(self, bbox_preds: Tensor, reg_targets: Tuple[list,int], batch_gt_instances: InstanceList, batch_img_metas: List[dict]) -> Tuple[Tensor]: """Loss function for outputs from a single decoder layer of a single feature level. Args: cls_scores (Tensor): Box score logits from a single decoder layer for all images, has shape (bs, num_queries, cls_out_channels). bbox_preds (Tensor): Sigmoid outputs from a single decoder layer for all images, with normalized coordinate (cx, cy, w, h) and shape (bs, num_queries, 4). batch_gt_instances (list[:obj:`InstanceData`]): Batch of gt_instance. It usually includes ``bboxes`` and ``labels`` attributes. batch_img_metas (list[dict]): Meta information of each image, e.g., image size, scaling factor, etc. Returns: Tuple[Tensor]: A tuple including `loss_cls`, `loss_box` and `loss_iou`. """ num_imgs = bbox_preds.size(0) (bbox_targets_list, bbox_weights_list, num_total_pos, num_total_neg) = reg_targets bbox_targets = torch.cat(bbox_targets_list, 0) bbox_weights = torch.cat(bbox_weights_list, 0) # Compute the average number of gt boxes across all gpus, for # normalization purposes num_total_pos = bbox_targets.new_tensor([num_total_pos]) num_total_pos = torch.clamp(reduce_mean(num_total_pos), min=1).item() # construct factors used for rescale bboxes factors = [] for img_meta, bbox_pred in zip(batch_img_metas, bbox_preds): img_h, img_w, = img_meta['img_shape'] factor = bbox_pred.new_tensor([img_w, img_h, img_w, img_h]).unsqueeze(0).repeat( bbox_pred.size(0), 1) factors.append(factor) factors = torch.cat(factors, 0) # DETR regress the relative position of boxes (cxcywh) in the image, # thus the learning target is normalized by the image size. So here # we need to re-scale them for calculating IoU loss bbox_preds = bbox_preds.reshape(-1, 4) bboxes = bbox_cxcywh_to_xyxy(bbox_preds) * factors bboxes_gt = bbox_cxcywh_to_xyxy(bbox_targets) * factors # regression IoU loss, defaultly GIoU loss loss_iou = self.loss_iou( bboxes, bboxes_gt, bbox_weights, avg_factor=num_total_pos) # regression L1 loss loss_bbox = self.loss_bbox( bbox_preds, bbox_targets, bbox_weights, avg_factor=num_total_pos) return loss_bbox, loss_iou def get_targets(self, cls_scores_list: List[Tensor], bbox_preds_list: List[Tensor], batch_gt_instances: InstanceList, batch_img_metas: List[dict], with_neg_cls:bool=True, assigner_type:str = None) -> tuple: """Compute regression and classification targets for a batch image. Outputs from a single decoder layer of a single feature level are used. Args: cls_scores_list (list[Tensor]): Box score logits from a single decoder layer for each image, has shape [num_queries, cls_out_channels]. bbox_preds_list (list[Tensor]): Sigmoid outputs from a single decoder layer for each image, with normalized coordinate (cx, cy, w, h) and shape [num_queries, 4]. batch_gt_instances (list[:obj:`InstanceData`]): Batch of gt_instance. It usually includes ``bboxes`` and ``labels`` attributes. batch_img_metas (list[dict]): Meta information of each image, e.g., image size, scaling factor, etc. Returns: tuple: a tuple containing the following targets. - labels_list (list[Tensor]): Labels for all images. - label_weights_list (list[Tensor]): Label weights for all images. - bbox_targets_list (list[Tensor]): BBox targets for all images. - bbox_weights_list (list[Tensor]): BBox weights for all images. - num_total_pos (int): Number of positive samples in all images. - num_total_neg (int): Number of negative samples in all images. """ (labels_list, label_weights_list, bbox_targets_list, bbox_weights_list, pos_inds_list, neg_inds_list) = multi_apply(self._get_targets_single, cls_scores_list, bbox_preds_list, batch_gt_instances, batch_img_metas, with_neg_cls=with_neg_cls, assigner_type= assigner_type) num_total_pos = sum((inds.numel() for inds in pos_inds_list)) num_total_neg = sum((inds.numel() for inds in neg_inds_list)) if not with_neg_cls: num_total_neg = 0 return (labels_list, label_weights_list, bbox_targets_list, bbox_weights_list, num_total_pos, num_total_neg) def get_targets_bbox(self, cls_scores_list: List[Tensor], bbox_preds_list: List[Tensor], batch_gt_instances: InstanceList, batch_img_metas: List[dict]) -> tuple: """Compute regression and classification targets for a batch image. Outputs from a single decoder layer of a single feature level are used. Args: cls_scores_list (list[Tensor]): Box score logits from a single decoder layer for each image, has shape [num_queries, cls_out_channels]. bbox_preds_list (list[Tensor]): Sigmoid outputs from a single decoder layer for each image, with normalized coordinate (cx, cy, w, h) and shape [num_queries, 4]. batch_gt_instances (list[:obj:`InstanceData`]): Batch of gt_instance. It usually includes ``bboxes`` and ``labels`` attributes. batch_img_metas (list[dict]): Meta information of each image, e.g., image size, scaling factor, etc. Returns: tuple: a tuple containing the following targets. - labels_list (list[Tensor]): Labels for all images. - label_weights_list (list[Tensor]): Label weights for all images. - bbox_targets_list (list[Tensor]): BBox targets for all images. - bbox_weights_list (list[Tensor]): BBox weights for all images. - num_total_pos (int): Number of positive samples in all images. - num_total_neg (int): Number of negative samples in all images. """ (bbox_targets_list, bbox_weights_list, pos_inds_list, neg_inds_list) = multi_apply(self._get_targets_single_bbox, cls_scores_list, bbox_preds_list, batch_gt_instances, batch_img_metas) num_total_pos = sum((inds.numel() for inds in pos_inds_list)) num_total_neg = sum((inds.numel() for inds in neg_inds_list)) return (bbox_targets_list, bbox_weights_list, num_total_pos, num_total_neg) def _loss_dn_single(self, dn_bbox_preds: Tensor, reg_targets: Tuple[list, int], batch_gt_instances: InstanceList, batch_img_metas: List[dict], dn_meta: Dict[str, int]) -> Tuple[Tensor]: """Denoising loss for outputs from a single decoder layer. Args: dn_cls_scores (Tensor): Classification scores of a single decoder layer in denoising part, has shape (bs, num_denoising_queries, cls_out_channels). dn_bbox_preds (Tensor): Regression outputs of a single decoder layer in denoising part. Each is a 4D-tensor with normalized coordinate format (cx, cy, w, h) and has shape (bs, num_denoising_queries, 4). batch_gt_instances (list[:obj:`InstanceData`]): Batch of gt_instance. It usually includes ``bboxes`` and ``labels`` attributes. batch_img_metas (list[dict]): Meta information of each image, e.g., image size, scaling factor, etc. dn_meta (Dict[str, int]): The dictionary saves information about group collation, including 'num_denoising_queries' and 'num_denoising_groups'. It will be used for split outputs of denoising and matching parts and loss calculation. Returns: Tuple[Tensor]: A tuple including `loss_cls`, `loss_box` and `loss_iou`. """ (labels_list, label_weights_list, bbox_targets_list, bbox_weights_list,num_total_pos) = reg_targets bbox_targets = torch.cat(bbox_targets_list, 0) bbox_weights = torch.cat(bbox_weights_list, 0) # Compute the average number of gt boxes across all gpus, for # normalization purposes num_total_pos = dn_bbox_preds.new_tensor([num_total_pos]) num_total_pos = torch.clamp(reduce_mean(num_total_pos), min=1).item() # construct factors used for rescale bboxes factors = [] for img_meta, bbox_pred in zip(batch_img_metas, dn_bbox_preds): img_h, img_w = img_meta['img_shape'] factor = bbox_pred.new_tensor([img_w, img_h, img_w, img_h]).unsqueeze(0).repeat( bbox_pred.size(0), 1) factors.append(factor) factors = torch.cat(factors) # DETR regress the relative position of boxes (cxcywh) in the image, # thus the learning target is normalized by the image size. So here # we need to re-scale them for calculating IoU loss bbox_preds = dn_bbox_preds.reshape(-1, 4) bboxes = bbox_cxcywh_to_xyxy(bbox_preds) * factors bboxes_gt = bbox_cxcywh_to_xyxy(bbox_targets) * factors # regression IoU loss, defaultly GIoU loss loss_iou = self.loss_iou( bboxes, bboxes_gt, bbox_weights, avg_factor=num_total_pos) # regression L1 loss loss_bbox = self.loss_bbox( bbox_preds, bbox_targets, bbox_weights, avg_factor=num_total_pos) return loss_bbox, loss_iou def get_dn_targets(self, batch_gt_instances: InstanceList, batch_img_metas: dict, dn_meta: Dict[str, int]) -> tuple: """Get targets in denoising part for a batch of images. Args: batch_gt_instances (list[:obj:`InstanceData`]): Batch of gt_instance. It usually includes ``bboxes`` and ``labels`` attributes. batch_img_metas (list[dict]): Meta information of each image, e.g., image size, scaling factor, etc. dn_meta (Dict[str, int]): The dictionary saves information about group collation, including 'num_denoising_queries' and 'num_denoising_groups'. It will be used for split outputs of denoising and matching parts and loss calculation. Returns: tuple: a tuple containing the following targets. - labels_list (list[Tensor]): Labels for all images. - label_weights_list (list[Tensor]): Label weights for all images. - bbox_targets_list (list[Tensor]): BBox targets for all images. - bbox_weights_list (list[Tensor]): BBox weights for all images. - num_total_pos (int): Number of positive samples in all images. - num_total_neg (int): Number of negative samples in all images. """ (labels_list, label_weights_list, bbox_targets_list, bbox_weights_list,pos_inds_list) = multi_apply( self._get_dn_targets_single, batch_gt_instances, batch_img_metas, dn_meta=dn_meta) num_total_pos = sum((inds.numel() for inds in pos_inds_list)) return (labels_list, label_weights_list, bbox_targets_list, bbox_weights_list, num_total_pos,) def _get_dn_targets_single(self, gt_instances: InstanceData, img_meta: dict, dn_meta: Dict[str, int]) -> tuple: """Get targets in denoising part for one image. Args: gt_instances (:obj:`InstanceData`): Ground truth of instance annotations. It should includes ``bboxes`` and ``labels`` attributes. img_meta (dict): Meta information for one image. dn_meta (Dict[str, int]): The dictionary saves information about group collation, including 'num_denoising_queries' and 'num_denoising_groups'. It will be used for split outputs of denoising and matching parts and loss calculation. Returns: tuple[Tensor]: a tuple containing the following for one image. - labels (Tensor): Labels of each image. - label_weights (Tensor]): Label weights of each image. - bbox_targets (Tensor): BBox targets of each image. - bbox_weights (Tensor): BBox weights of each image. - pos_inds (Tensor): Sampled positive indices for each image. - neg_inds (Tensor): Sampled negative indices for each image. """ gt_bboxes = gt_instances.bboxes gt_labels = gt_instances.labels num_groups = dn_meta['num_denoising_groups'] num_denoising_queries = dn_meta['num_denoising_queries'] num_queries_each_group = int(num_denoising_queries / num_groups) device = gt_bboxes.device if len(gt_labels) > 0: t = torch.arange(len(gt_labels), dtype=torch.long, device=device) t = t.unsqueeze(0).repeat(num_groups, 1) pos_assigned_gt_inds = t.flatten() pos_inds = torch.arange(num_groups, dtype=torch.long, device=device) pos_inds = pos_inds.unsqueeze(1) * num_queries_each_group + t pos_inds = pos_inds.flatten() else: pos_inds = pos_assigned_gt_inds = \ gt_bboxes.new_tensor([], dtype=torch.long) # label targets labels = gt_bboxes.new_full((num_denoising_queries, ), self.num_classes, dtype=torch.long) labels[pos_inds] = gt_labels[pos_assigned_gt_inds] label_weights = gt_bboxes.new_zeros(num_denoising_queries) label_weights[pos_inds] = 1.0 # bbox targets bbox_targets = torch.zeros(num_denoising_queries, 4, device=device) bbox_weights = torch.zeros(num_denoising_queries, 4, device=device) bbox_weights[pos_inds] = 1.0 img_h, img_w = img_meta['img_shape'] # DETR regress the relative position of boxes (cxcywh) in the image. # Thus the learning target should be normalized by the image size, also # the box format should be converted from defaultly x1y1x2y2 to cxcywh. factor = gt_bboxes.new_tensor([img_w, img_h, img_w, img_h]).unsqueeze(0) gt_bboxes_normalized = gt_bboxes / factor gt_bboxes_targets = bbox_xyxy_to_cxcywh(gt_bboxes_normalized) bbox_targets[pos_inds] = gt_bboxes_targets.repeat([num_groups, 1]) return (labels, label_weights,bbox_targets, bbox_weights, pos_inds) @staticmethod def split_outputs(all_layers_cls_scores: Tensor, all_layers_bbox_preds: Tensor, dn_meta: Dict[str, int]) -> Tuple[Tensor]: """Split outputs of the denoising part and the matching part. For the total outputs of `num_queries_total` length, the former `num_denoising_queries` outputs are from denoising queries, and the rest `num_matching_queries` ones are from matching queries, where `num_queries_total` is the sum of `num_denoising_queries` and `num_matching_queries`. Args: all_layers_cls_scores (Tensor): Classification scores of all decoder layers, has shape (num_decoder_layers, bs, num_queries_total, cls_out_channels). all_layers_bbox_preds (Tensor): Regression outputs of all decoder layers. Each is a 4D-tensor with normalized coordinate format (cx, cy, w, h) and has shape (num_decoder_layers, bs, num_queries_total, 4). dn_meta (Dict[str, int]): The dictionary saves information about group collation, including 'num_denoising_queries' and 'num_denoising_groups'. Returns: Tuple[Tensor]: a tuple containing the following outputs. - all_layers_matching_cls_scores (Tensor): Classification scores of all decoder layers in matching part, has shape (num_decoder_layers, bs, num_matching_queries, cls_out_channels). - all_layers_matching_bbox_preds (Tensor): Regression outputs of all decoder layers in matching part. Each is a 4D-tensor with normalized coordinate format (cx, cy, w, h) and has shape (num_decoder_layers, bs, num_matching_queries, 4). - all_layers_denoising_cls_scores (Tensor): Classification scores of all decoder layers in denoising part, has shape (num_decoder_layers, bs, num_denoising_queries, cls_out_channels). - all_layers_denoising_bbox_preds (Tensor): Regression outputs of all decoder layers in denoising part. Each is a 4D-tensor with normalized coordinate format (cx, cy, w, h) and has shape (num_decoder_layers, bs, num_denoising_queries, 4). """ if dn_meta is not None: num_denoising_queries = dn_meta['num_denoising_queries'] all_layers_denoising_cls_scores = \ all_layers_cls_scores[:,:, : num_denoising_queries, :] all_layers_denoising_bbox_preds = \ all_layers_bbox_preds[:, :, : num_denoising_queries, :] all_layers_matching_cls_scores = \ all_layers_cls_scores[:, :, num_denoising_queries:, :] all_layers_matching_bbox_preds = \ all_layers_bbox_preds[:, :, num_denoising_queries:, :] else: all_layers_denoising_cls_scores = None all_layers_denoising_bbox_preds = None all_layers_matching_cls_scores = all_layers_cls_scores all_layers_matching_bbox_preds = all_layers_bbox_preds return (all_layers_matching_cls_scores, all_layers_matching_bbox_preds, all_layers_denoising_cls_scores, all_layers_denoising_bbox_preds) @staticmethod def split_outputsv1(all_layers_cls_scores: Tensor, all_layers_bbox_preds: Tensor, dn_meta: Dict[str, int]) -> Tuple[Tensor]: """Split outputs of the denoising part and the matching part. For the total outputs of `num_queries_total` length, the former `num_denoising_queries` outputs are from denoising queries, and the rest `num_matching_queries` ones are from matching queries, where `num_queries_total` is the sum of `num_denoising_queries` and `num_matching_queries`. Args: all_layers_cls_scores (Tensor): Classification scores of all decoder layers, has shape (num_decoder_layers, bs, num_queries_total, cls_out_channels). all_layers_bbox_preds (Tensor): Regression outputs of all decoder layers. Each is a 4D-tensor with normalized coordinate format (cx, cy, w, h) and has shape (num_decoder_layers, bs, num_queries_total, 4). dn_meta (Dict[str, int]): The dictionary saves information about group collation, including 'num_denoising_queries' and 'num_denoising_groups'. Returns: Tuple[Tensor]: a tuple containing the following outputs. - all_layers_matching_cls_scores (Tensor): Classification scores of all decoder layers in matching part, has shape (num_decoder_layers, bs, num_matching_queries, cls_out_channels). - all_layers_matching_bbox_preds (Tensor): Regression outputs of all decoder layers in matching part. Each is a 4D-tensor with normalized coordinate format (cx, cy, w, h) and has shape (num_decoder_layers, bs, num_matching_queries, 4). - all_layers_denoising_cls_scores (Tensor): Classification scores of all decoder layers in denoising part, has shape (num_decoder_layers, bs, num_denoising_queries, cls_out_channels). - all_layers_denoising_bbox_preds (Tensor): Regression outputs of all decoder layers in denoising part. Each is a 4D-tensor with normalized coordinate format (cx, cy, w, h) and has shape (num_decoder_layers, bs, num_denoising_queries, 4). """ if dn_meta is not None: num_denoising_queries = dn_meta['num_denoising_queries'] all_layers_denoising_cls_scores = \ all_layers_cls_scores[:, : num_denoising_queries, :] all_layers_denoising_bbox_preds = \ all_layers_bbox_preds[:, :, : num_denoising_queries, :] all_layers_matching_cls_scores = \ all_layers_cls_scores[:, num_denoising_queries:, :] all_layers_matching_bbox_preds = \ all_layers_bbox_preds[:, :, num_denoising_queries:, :] else: all_layers_denoising_cls_scores = None all_layers_denoising_bbox_preds = None all_layers_matching_cls_scores = all_layers_cls_scores all_layers_matching_bbox_preds = all_layers_bbox_preds return (all_layers_matching_cls_scores, all_layers_matching_bbox_preds, all_layers_denoising_cls_scores, all_layers_denoising_bbox_preds) def predict(self, hidden_states: Tensor, references: List[Tensor], centers: Tensor, center_scores: Tensor, topk_centers_scores: Tensor, cls_feats: Tensor, batch_data_samples: SampleList, rescale: bool = True) -> InstanceList: """Perform forward propagation and loss calculation of the detection head on the queries of the upstream network. Args: hidden_states (Tensor): Hidden states output from each decoder layer, has shape (num_decoder_layers, num_queries, bs, dim). references (list[Tensor]): List of the reference from the decoder. The first reference is the `init_reference` (initial) and the other num_decoder_layers(6) references are `inter_references` (intermediate). The `init_reference` has shape (bs, num_queries, 4) when `as_two_stage` of the detector is `True`, otherwise (bs, num_queries, 2). Each `inter_reference` has shape (bs, num_queries, 4) when `with_box_refine` of the detector is `True`, otherwise (bs, num_queries, 2). The coordinates are arranged as (cx, cy) when the last dimension is 2, and (cx, cy, w, h) when it is 4. batch_data_samples (list[:obj:`DetDataSample`]): The Data Samples. It usually includes information such as `gt_instance`, `gt_panoptic_seg` and `gt_sem_seg`. rescale (bool, optional): If `True`, return boxes in original image space. Defaults to `True`. Returns: list[obj:`InstanceData`]: Detection results of each image after the post process. """ batch_img_metas = [ data_samples.metainfo for data_samples in batch_data_samples ] outs = self(hidden_states, references, cls_feats) predictions = self.predict_by_feat( *outs, batch_img_metas=batch_img_metas, rescale=rescale) return predictions def predict_by_feat(self, cls_scores: Tensor, inputs_coords: Tensor, all_layers_cls_scores: Tensor, all_layers_bbox_preds: Tensor, batch_img_metas: List[Dict], rescale: bool = False) -> InstanceList: """Transform a batch of output features extracted from the head into bbox results. Args: all_layers_cls_scores (Tensor): Classification scores of all decoder layers, has shape (num_decoder_layers, bs, num_queries, cls_out_channels). all_layers_bbox_preds (Tensor): Regression outputs of all decoder layers. Each is a 4D-tensor with normalized coordinate format (cx, cy, w, h) and shape (num_decoder_layers, bs, num_queries, 4) with the last dimension arranged as (cx, cy, w, h). batch_img_metas (list[dict]): Meta information of each image. rescale (bool, optional): If `True`, return boxes in original image space. Default `False`. Returns: list[obj:`InstanceData`]: Detection results of each image after the post process. """ cls_scores = all_layers_cls_scores[-1] bbox_preds = all_layers_bbox_preds[-1] result_list = [] for img_id in range(len(batch_img_metas)): cls_score = cls_scores[img_id] bbox_pred = bbox_preds[img_id] img_meta = batch_img_metas[img_id] results = self._predict_by_feat_single(cls_score, bbox_pred, img_meta, rescale) result_list.append(results) return result_list def _predict_by_feat_single(self, cls_score: Tensor, bbox_pred: Tensor, img_meta: dict, rescale: bool = True) -> InstanceData: """Transform outputs from the last decoder layer into bbox predictions for each image. Args: cls_score (Tensor): Box score logits from the last decoder layer for each image. Shape [num_queries, cls_out_channels]. bbox_pred (Tensor): Sigmoid outputs from the last decoder layer for each image, with coordinate format (cx, cy, w, h) and shape [num_queries, 4]. img_meta (dict): Image meta info. rescale (bool): If True, return boxes in original image space. Default True. Returns: :obj:`InstanceData`: Detection results of each image after the post process. Each item usually contains following keys. - scores (Tensor): Classification scores, has a shape (num_instance, ) - labels (Tensor): Labels of bboxes, has a shape (num_instances, ). - bboxes (Tensor): Has a shape (num_instances, 4), the last dimension 4 arrange as (x1, y1, x2, y2). """ assert len(cls_score) == len(bbox_pred) # num_queries max_per_img = self.test_cfg.get('max_per_img', len(cls_score)) img_shape = img_meta['img_shape'] assert self.loss_cls.use_sigmoid cls_score = cls_score.sigmoid() # scores, indexes = cls_score.view(-1).topk(max_per_img) scores, indexes = torch.sort(cls_score.view(-1), descending=True) indexes = indexes[scores > self.iou_threshold] scores = scores[scores > self.iou_threshold] det_labels = indexes % self.num_classes bbox_index = torch.div(indexes, self.num_classes, rounding_mode='trunc') bbox_pred = bbox_pred[bbox_index] det_bboxes = bbox_cxcywh_to_xyxy(bbox_pred) det_bboxes[:, 0::2] = det_bboxes[:, 0::2] * img_shape[1] det_bboxes[:, 1::2] = det_bboxes[:, 1::2] * img_shape[0] det_bboxes[:, 0::2].clamp_(min=0, max=img_shape[1]) det_bboxes[:, 1::2].clamp_(min=0, max=img_shape[0]) if self.use_nms: if det_labels.numel() > 0: bboxes_scores, keep = batched_nms(det_bboxes, scores.contiguous(), det_labels, self.test_nms, class_agnostic=(not self.class_wise_nms)) if keep.numel() > max_per_img: bboxes_scores = bboxes_scores[:max_per_img] det_labels = det_labels[keep][:max_per_img] else: det_labels = det_labels[keep] det_bboxes = bboxes_scores[:, :-1] scores = bboxes_scores[:, -1] if self.pre_bboxes_round: det_bboxes = adjust_bbox_to_pixel(det_bboxes) if rescale: # assert img_meta.get('scale_factor') is not None # det_bboxes /= det_bboxes.new_tensor( # img_meta['scale_factor']).repeat((1, 2)) # rw by lzx if img_meta.get('scale_factor') is not None: det_bboxes /= det_bboxes.new_tensor( img_meta['scale_factor']).repeat((1, 2)) results = InstanceData() results.bboxes = det_bboxes results.scores = scores results.labels = det_labels return results