Hi everyone,
I trained keypointrcnn_resnet50_fpn from torchvision to predict 12 keypoints. The training phase was good but I got an error when doing inference.
Here is my training loop:
valid_loss_min = np.inf
for epoch in range(1, num_epochs+1):
epoch += epoch_resume
train_loss_per_batch = []
val_loss_per_batch = []
for images, targets in train_loader:
images = list(image.type(dtype).to(device) for image in images)
targets = [{k: v.type(dtype).to(device) for k, v in t.items()} for t in targets]
loss_dict = model(images, targets)
losses = loss_dict['loss_keypoint']
optimizer.zero_grad()
losses.backward()
optimizer.step()
train_loss_per_batch.append(losses.item())
# Validation
with torch.no_grad():
for images, targets in valid_loader:
images = list(image.type(dtype).to(device) for image in images)
targets = [{k: v.type(dtype).to(device) for k, v in t.items()} for t in targets]
loss_dict = model(images, targets)
losses = loss_dict['loss_keypoint']
val_loss_per_batch.append(losses.item())
avg_val_loss = np.mean(np.array(val_loss_per_batch))
avg_train_loss = np.mean(np.array(train_loss_per_batch))
print(f"Epoch [{epoch}]\t- Train Loss: {avg_train_loss}\t- Validation Loss: {avg_val_loss} ")
if avg_val_loss <= valid_loss_min:
print(f"\nValidation loss decreased ({valid_loss_min} --> {avg_val_loss}). Saving model ...\n")
torch.save({
'epoch': epoch,
'model_state_dict': model.state_dict(),
'optimizer_state_dict': optimizer.state_dict(),
'best_loss': avg_val_loss
}, saved_model_path)
valid_loss_min = avg_val_loss
And here is the code for inference:
model, device, dtype = load_model(model_path='models/model_resume_fot.pt', nbkpts=12)
img_name = 'img_test/1.png'
img = cv2.imread(img_name)
new_h, new_w = img.shape[:2]
img_resized = cv2.resize(img, (448, 448))
img_pil = T.ToPILImage()(img_resized).convert('RGB')
im_tensor = T.ToTensor()(img_pil)
with torch.no_grad():
prediction = model([im_tensor.to(device)])
And got this error:
Traceback (most recent call last):
File "test_inf_image.py", line 75, in <module>
prediction = model([im_tensor.to(device)])
File "/Users/medhyvinceslas/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 493, in __call__
result = self.forward(*input, **kwargs)
File "/Users/medhyvinceslas/anaconda3/lib/python3.7/site-packages/torchvision-0.3.0a0+427633a-py3.7-macosx-10.7-x86_64.egg/torchvision/models/detection/generalized_rcnn.py", line 52, in forward
File "/Users/medhyvinceslas/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 493, in __call__
result = self.forward(*input, **kwargs)
File "/Users/medhyvinceslas/anaconda3/lib/python3.7/site-packages/torchvision-0.3.0a0+427633a-py3.7-macosx-10.7-x86_64.egg/torchvision/models/detection/roi_heads.py", line 598, in forward
File "/Users/medhyvinceslas/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 493, in __call__
result = self.forward(*input, **kwargs)
File "/Users/medhyvinceslas/anaconda3/lib/python3.7/site-packages/torchvision-0.3.0a0+427633a-py3.7-macosx-10.7-x86_64.egg/torchvision/models/detection/keypoint_rcnn.py", line 253, in forward
File "/Users/medhyvinceslas/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 493, in __call__
result = self.forward(*input, **kwargs)
File "/Users/medhyvinceslas/anaconda3/lib/python3.7/site-packages/torch/nn/modules/conv.py", line 796, in forward
output_padding, self.groups, self.dilation)
RuntimeError: invalid argument 2: non-empty 3D or 4D input tensor expected but got: [0 x 512 x 14 x 14] at /Users/distiller/project/conda/conda-bld/pytorch_1556653464916/work/aten/src/THNN/generic/SpatialFullDilatedConvolution.c:44
I’m sure that my image exists, so I don’t know what happens here. Can you help me?
Thank you.
Medhy
