about gradcam

[5nmPlus]

          I also want to ask this question.

When I tried to use tools/misc/vis_cam.py(--target-layers: head.transformer.decoder.layers[2].attentions[0].attn),I encountered the following problem:

Traceback (most recent call last):
File "/SimVG/tools/misc/vis_cam.py", line 224, in
main()
File "SimVG/tools/misc/vis_cam.py", line 167, in main
grayscale_cam = det_cam_visualizer(image, targets=targets, aug_smooth=args.aug_smooth, eigen_smooth=args.eigen_smooth)
File "SimVG/simvg/utils/det_cam_visualizer.py", line 268, in call
return self.cam(img, targets, aug_smooth, eigen_smooth)[0, :]
File "/home/li/.conda/envs/simvg/lib/python3.10/site-packages/pytorch_grad_cam/base_cam.py", line 184, in call
return self.forward(input_tensor,
File "/home/li/.conda/envs/simvg/lib/python3.10/site-packages/pytorch_grad_cam/base_cam.py", line 74, in forward
outputs = self.activations_and_grads(input_tensor)
File "SimVG/simvg/utils/det_cam_visualizer.py", line 430, in call
return self.model(x)
File "SimVG/simvg/utils/det_cam_visualizer.py", line 162, in call
loss = self.detector(return_loss=True, **self.input_data)[0]
File "/home/li/.conda/envs/simvg/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "/home/li/.conda/envs/simvg/lib/python3.10/site-packages/mmcv/runner/fp16_utils.py", line 119, in new_func
return old_func(*args, **kwargs)
File "/SimVG/simvg/models/det_seg/base.py", line 25, in forward
return self.forward_train(img, ref_expr_inds, img_metas, **kwargs)
File "SimVG/simvg/models/det_seg/mix_detr_mb.py", line 54, in forward_train
losses_dict, output = self.head.forward_train(
File "/SimVG/simvg/models/heads/tgqs_kd_detr_head/tgqs_kd_detr_head.py", line 458, in forward_train
output = self.forward_general(x_mm, img_metas, cls_feat=cls_feat, text_feat=text_feat, text_mask=text_mask)
File "SimVG/simvg/models/heads/tgqs_kd_detr_head/tgqs_kd_detr_head.py", line 425, in forward_general
hidden_states, _ = self.transformer(x_mm, img_masks, query_embed, pos_embed)
File "/home/li/.conda/envs/simvg/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "/SimVG/simvg/models/heads/tgqs_kd_detr_head/transformer.py", line 225, in forward
decoder_output = self.decoder(
File "/home/li/.conda/envs/simvg/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "/SimVG/simvg/models/heads/tgqs_kd_detr_head/transformer.py", line 168, in forward
query = layer(
File "/home/li/.conda/envs/simvg/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "/SimVG/tools/detrex/detrex/layers/transformer.py", line 155, in forward
query = self.attentions[attn_index](
File "/home/li/.conda/envs/simvg/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "SimVG/tools/detrex/detrex/layers/attention.py", line 130, in forward
out = self.attn(
File "/home/li/.conda/envs/simvg/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1547, in _call_impl
hook_result = hook(self, args, result)
File "SimVG/simvg/utils/det_cam_visualizer.py", line 411, in save_activation
activation = self.reshape_transform(activation[0])
File "/SimVG/simvg/utils/det_cam_visualizer.py", line 69, in reshape_transform
activations.append(torch.nn.functional.interpolate(torch.abs(feat), max_shape, mode="bilinear"))
File "/home/li/.conda/envs/simvg/lib/python3.10/site-packages/torch/nn/functional.py", line 3959, in interpolate
return torch._C._nn.upsample_bilinear2d(input, output_size, align_corners, scale_factors)
RuntimeError: Non-empty 4D data tensor expected but got a tensor with sizes [1, 0, 20, 20]

Originally posted by @5nmPlus in #9 (comment)

[5nmPlus]

When I tried to use tools/misc/vis_cam.py(--target-layers: head.transformer.decoder.layers[2].attentions[0].attn),I encountered the following problem:

RuntimeError: Non-empty 4D data tensor expected but got a tensor with sizes [1, 0, 20, 20]

Originally posted by @5nmPlus in #9 (comment)

When I try to set -target-layers to vis_enc.beit3.encoder.layers[11]. It worked, but the result was obviously wrong