Vision: engine.py error while following tutorial

[EMCP]

📚 Documentation

I have found this library via the examples at https://pytorch.org/tutorials/intermediate/torchvision_tutorial.html

I ran the google colab and that successfully finished.. however when I go to copy master from here and use it locally I am getting an error..

Engine.py line

Error

` train_one_epoch(model, optimizer, training_data_loader, device, epoch, print_freq=model_conf["hyperParameters"]["display"]) File "/home/emcp/Dev/git/EMCP/faster-rcnn-torchvision/model_components/model/engine.py", line 27, in train_one_epoch images = list(image.to(device) for image in images) File "/home/emcp/Dev/git/EMCP/faster-rcnn-torchvision/model_components/model/engine.py", line 27, in <genexpr> images = list(image.to(device) for image in images) AttributeError: 'Image' object has no attribute 'to'
seems to be an image did not load perhaps ?

[EMCP]

When I look at my DataLoader it seems to be giving correct items

The image is COCO annotated just FYI

(<PIL.Image.Image image mode=RGB size=2880x1880 at 0x7F06E2CCAAD0>, [{'id': 847, 'image_id': 3135, 'category_id': 6, 'segmentation': [[90.0, 36.0, 2874.8, 35.7, 2880.0, 40.1, 2880.0, 1795.3, 2875.3, 1800.0, 3.6, 1799.7, 0.4, 1795.5, 0.2, 41.2, 6.1, 35.7, 90.0, 36.1]], 'area': 5080276, 'bbox': [0.0, 36.0, 2880.0, 1764.0], 'iscrowd': False, 'color': '#d5e47c', 'metadata': {}}, {'id': 848, 'image_id': 3135, 'category_id': 7, 'segmentation': [[666.9, 1799.9, 2239.9, 1800.0, 2245.7, 1805.4, 2246.0, 1879.8, 634.8, 1879.9, 634.2, 1805.9, 639.4, 1799.9, 667.3, 1800.2]], 'area': 128892, 'bbox': [634.0, 1800.0, 1612.0, 80.0], 'iscrowd': False, 'color': '#6c3ee0', 'metadata': {}}])

[fmassa]

@EMCP You need to feed a torch.Tensor to the model. Your Dataset is returning a PIL Image, that's why it's not working. We convert the PIL Image into a torch Tensor in the dataset (via the transform.

[EMCP]

big thank you.. Feels like I'm almost there then with it

[kotchwane]

I'm having the exact same issue here... Could you be more precise about what needs to be changed? I checked 10 times, I did not miss one step from the tutorial.

[kotchwane]

Okay, I found it. While doing the tutorial, I had commented these lines, inside PennFudanDataSet, because they created problems for early testing :

if self.transforms is not None:
            img, target = self.transforms(img, target)

Un-commenting these solved the issue (and, of course, led to a new one).

What I had not realized earlier in the tutorial, and led me to wrongly comment those two lines, is that you can easily test the PennFudanDataSet by calling it with None as a second argument:
dataset = PennFudanDataset('PennFudanPed', None)