detr_prof

Timer unit: 1e-06 s

Total time: 0.112193 s
File: /home/zhiyilai/.cache/torch/hub/facebookresearch_detr_master/models/detr.py
Function: forward at line 44

Line #      Hits         Time  Per Hit   % Time  Line Contents
==============================================================
    44                                               def forward(self, samples: NestedTensor):
    45                                                   """ The forward expects a NestedTensor, which consists of:
    46                                                          - samples.tensor: batched images, of shape [batch_size x 3 x H x W]
    47                                                          - samples.mask: a binary mask of shape [batch_size x H x W], containing 1 on padded pixels
    48                                           
    49                                                       It returns a dict with the following elements:
    50                                                          - "pred_logits": the classification logits (including no-object) for all queries.
    51                                                                           Shape= [batch_size x num_queries x (num_classes + 1)]
    52                                                          - "pred_boxes": The normalized boxes coordinates for all queries, represented as
    53                                                                          (center_x, center_y, height, width). These values are normalized in [0, 1],
    54                                                                          relative to the size of each individual image (disregarding possible padding).
    55                                                                          See PostProcess for information on how to retrieve the unnormalized bounding box.
    56                                                          - "aux_outputs": Optional, only returned when auxilary losses are activated. It is a list of
    57                                                                           dictionnaries containing the two above keys for each decoder layer.
    58                                                   """
    59         1          6.0      6.0      0.0          if isinstance(samples, (list, torch.Tensor)):
    60         1        974.0    974.0      0.9              samples = nested_tensor_from_tensor_list(samples)
    61         1      66212.0  66212.0     59.0          features, pos = self.backbone(samples)
    62                                           
    63         1          7.0      7.0      0.0          src, mask = features[-1].decompose()
    64         1          1.0      1.0      0.0          assert mask is not None
    65         1      44287.0  44287.0     39.5          hs = self.transformer(self.input_proj(src), mask, self.query_embed.weight, pos[-1])[0]
    66                                           
    67         1        145.0    145.0      0.1          outputs_class = self.class_embed(hs)
    68         1        541.0    541.0      0.5          outputs_coord = self.bbox_embed(hs).sigmoid()
    69         1         17.0     17.0      0.0          out = {'pred_logits': outputs_class[-1], 'pred_boxes': outputs_coord[-1]}
    70         1          2.0      2.0      0.0          if self.aux_loss:
    71                                                       out['aux_outputs'] = self._set_aux_loss(outputs_class, outputs_coord)
    72         1          1.0      1.0      0.0          return out