Question: AssertionError

[dgleba]

Describe the question

I Doubt it's a bug. Just a question.

I am new to pytorch etc. I get this error with my images. but not with other images I got from a public dataset.

AssertionError:

It doesn't say much. There is nothing after the colon. It simply says AssertionError: -- that's it -- nothing more. How can I figure out what I am doing wrong?

I wonder if it has to do with my images. They are all the same size. I hoped the default transformation would be enough.

Code and Data Share the code that caused the error as well as an example image/label from your dataset.

I have started with your colab notebook at: https://colab.research.google.com/drive/1ISaTV5F-7b4i2QqtjTa7ToDPQ2k8qEe0 I have used a few of my images.

code:

losses = model.fit(loader, val_dataset, epochs=3, verbose=True)

output:

It looks like you're training your model on a CPU. Consider switching to a GPU; otherwise, this method can take hours upon hours or even days to finish. For more information, see https://detecto.readthedocs.io/en/latest/usage/quickstart.html#technical-requirements
Epoch 1 of 3
Begin iterating over training dataset
  0%|          | 0/15 [00:00<?, ?it/s]
---------------------------------------------------------------------------
AssertionError                            Traceback (most recent call last)
/tmp/ipykernel_77/1923109836.py in <cell line: 13>()
     11 # Train the model! This step can take a while, so make sure you
     12 # the GPU is turned on in Edit -> Notebook settings
---> 13 losses = model.fit(loader, val_dataset, epochs=3, verbose=True)
     14 
     15 # Plot the accuracy over time

~/.conda/envs/default/lib/python3.9/site-packages/detecto/core.py in fit(self, dataset, val_dataset, epochs, learning_rate, momentum, weight_decay, gamma, lr_step_size, verbose)
    521                 # Calculate the model's loss (i.e. how well it does on the current
    522                 # image and target, with a lower loss being better)
--> 523                 loss_dict = self._model(images, targets)
    524                 total_loss = sum(loss for loss in loss_dict.values())
    525 

~/.conda/envs/default/lib/python3.9/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1108         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1109                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1110             return forward_call(*input, **kwargs)
   1111         # Do not call functions when jit is used
   1112         full_backward_hooks, non_full_backward_hooks = [], []

~/.conda/envs/default/lib/python3.9/site-packages/torchvision/models/detection/generalized_rcnn.py in forward(self, images, targets)
     97             features = OrderedDict([("0", features)])
     98         proposals, proposal_losses = self.rpn(images, features, targets)
---> 99         detections, detector_losses = self.roi_heads(features, proposals, images.image_sizes, targets)
    100         detections = self.transform.postprocess(detections, images.image_sizes, original_image_sizes)  # type: ignore[operator]
    101 

~/.conda/envs/default/lib/python3.9/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1108         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1109                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1110             return forward_call(*input, **kwargs)
   1111         # Do not call functions when jit is used
   1112         full_backward_hooks, non_full_backward_hooks = [], []

~/.conda/envs/default/lib/python3.9/site-packages/torchvision/models/detection/roi_heads.py in forward(self, features, proposals, image_shapes, targets)
    749             matched_idxs = None
    750 
--> 751         box_features = self.box_roi_pool(features, proposals, image_shapes)
    752         box_features = self.box_head(box_features)
    753         class_logits, box_regression = self.box_predictor(box_features)

~/.conda/envs/default/lib/python3.9/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1108         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1109                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1110             return forward_call(*input, **kwargs)
   1111         # Do not call functions when jit is used
   1112         full_backward_hooks, non_full_backward_hooks = [], []

~/.conda/envs/default/lib/python3.9/site-packages/torchvision/ops/poolers.py in forward(self, x, boxes, image_shapes)
    325         x_filtered = _filter_input(x, self.featmap_names)
    326         if self.scales is None or self.map_levels is None:
--> 327             self.scales, self.map_levels = _setup_scales(
    328                 x_filtered, image_shapes, self.canonical_scale, self.canonical_level
    329             )

~/.conda/envs/default/lib/python3.9/site-packages/torchvision/ops/poolers.py in _setup_scales(features, image_shapes, canonical_scale, canonical_level)
    121     original_input_shape = (max_x, max_y)
    122 
--> 123     scales = [_infer_scale(feat, original_input_shape) for feat in features]
    124     # get the levels in the feature map by leveraging the fact that the network always
    125     # downsamples by a factor of 2 at each level.

~/.conda/envs/default/lib/python3.9/site-packages/torchvision/ops/poolers.py in <listcomp>(.0)
    121     original_input_shape = (max_x, max_y)
    122 
--> 123     scales = [_infer_scale(feat, original_input_shape) for feat in features]
    124     # get the levels in the feature map by leveraging the fact that the network always
    125     # downsamples by a factor of 2 at each level.

~/.conda/envs/default/lib/python3.9/site-packages/torchvision/ops/poolers.py in _infer_scale(feature, original_size)
    105         scale = 2 ** float(torch.tensor(approx_scale).log2().round())
    106         possible_scales.append(scale)
--> 107     assert possible_scales[0] == possible_scales[1]
    108     return possible_scales[0]
    109 

AssertionError: 

Environment:

• OS: Ubuntu 20.04 AWS Sagemaker studio lab
• Python version: Python 3.9.12
• Detecto version: 1.2.2 torch 1.11.0 torchvision 0.12.0

Additional context

All the files including code, images and xml files are here: https://github.com/dgleba/ml635e

Specifically: https://github.com/dgleba/ml635e/blob/main/ir.ipynb and https://github.com/dgleba/ml635e/tree/main/ir

The example I implemented in https://github.com/dgleba/ml635e/blob/main/e03cast.ipynb and its associated cast03 folder works OK.

[dgleba]

I was trying to narrow down the cause of the error.
I had the same error in both detecto and pytorch without detecto.

I resized the images from 260 x 7990 to a smaller size.
I resized the images in pytorch without detecto. The error went away and the training completed. I found the largest size that would work was 260 x 7500.

I figure that the error would be resolved in detecto as well.

It seems like the issue is something in pytorch or torch vision.