module has no attribute

[one-molamola]

Hello, I encountered the following problem when extracting the image features of the custom dataset, can you help me complete the code?

Traceback (most recent call last):
File "extract_features.py", line 82, in
pooled_feat = fasterRCNN(im_data, im_info, gt_boxes, num_boxes)
File "/root/miniconda3/envs/feature/lib/python3.6/site-packages/torch/nn/modules/module.py", line 491, in call
result = self.forward(*input, **kwargs)
File "/home/bottom-up-features-master/lib/model/faster_rcnn/faster_rcnn.py", line 73, in forward
pooled_feat = self.RCNN_roi_pool(base_feat, rois.view(-1, 5))
File "/root/miniconda3/envs/feature/lib/python3.6/site-packages/torch/nn/modules/module.py", line 491, in call
result = self.forward(*input, **kwargs)
File "/home/bottom-up-features-master/lib/model/roi_pooling/modules/roi_pool.py", line 14, in forward
return RoIPoolFunction(self.pooled_height, self.pooled_width, self.spatial_scale)(features, rois)
File "/home/bottom-up-features-master/lib/model/roi_pooling/functions/roi_pool.py", line 25, in forward
roi_pooling.roi_pooling_forward_cuda(ctx.pooled_height, ctx.pooled_width, ctx.spatial_scale,
AttributeError: module 'model.roi_pooling._ext.roi_pooling' has no attribute 'roi_pooling_forward_cuda'

[one-molamola]

I have solved this problem, but I have a new problem, I want to know what the extracted image feature dimension is 2048 (10-100)?

Because when I use the extracted image features to train the model, the following problem arises

RuntimeError: stack expects each tensor to be equal size, but got [100, 10] at entry 0 and [100, 18] at entry 1

[violetteshev]

Hi. It outputs 10-100 features (1 for each detected object), so the value will differ for each image.

[one-molamola]

So how do I use these features in the dataloader function? Because the following error was reported in the program

Traceback (most recent call last):
File "E:/vizwiz-vqa/mcan-vqa-master/run.py", line 155, in
execution.run(__C.RUN_MODE)
File "E:\vizwiz-vqa\mcan-vqa-master\core\exec.py", line 513, in run
self.train(self.dataset, self.dataset_eval)
File "E:\vizwiz-vqa\mcan-vqa-master\core\exec.py", line 149, in train
for step, (
File "C:\Users\KAI.conda\envs\fw\lib\site-packages\torch\utils\data\dataloader.py", line 521, in next
data = self._next_data()
File "C:\Users\KAI.conda\envs\fw\lib\site-packages\torch\utils\data\dataloader.py", line 561, in _next_data
data = self._dataset_fetcher.fetch(index) # may raise StopIteration
File "C:\Users\KAI.conda\envs\fw\lib\site-packages\torch\utils\data_utils\fetch.py", line 52, in fetch
return self.collate_fn(data)
File "C:\Users\KAI.conda\envs\fw\lib\site-packages\torch\utils\data_utils\collate.py", line 84, in default_collate
return [default_collate(samples) for samples in transposed]
File "C:\Users\KAI.conda\envs\fw\lib\site-packages\torch\utils\data_utils\collate.py", line 84, in
return [default_collate(samples) for samples in transposed]
File "C:\Users\KAI.conda\envs\fw\lib\site-packages\torch\utils\data_utils\collate.py", line 56, in default_collate
return torch.stack(batch, 0, out=out)
RuntimeError: stack expects each tensor to be equal size, but got [100, 14] at entry 0 and [100, 11] at entry 1

进程已结束,退出代码1

[one-molamola]

Can you provide some reference code to load these features?

[violetteshev]

You can pad all the features to the max size (100 in this case) with 0. That's what they do in the original MCAN model: code (in proc_img_feat function)