Running train.ipynb and pred_and_eval_ss.ipynb

[mmuneeburahman]

I am trying to reproduce the results from these two files. The train.ipynb is working fine and model is trained and saved. But when I load that model for prediction. I call the function: SemanticSegmentationLabels.from_predictions It gives an error.


NoteBook Link

Predicting:   0%|          | 0[/4](https://file+.vscode-resource.vscode-cdn.net/4) [00:02<?, ?it[/s](https://file+.vscode-resource.vscode-cdn.net/s)]
Output exceeds the [size limit](command:workbench.action.openSettings?%5B%22notebook.output.textLineLimit%22%5D). Open the full output data [in a text editor](command:workbench.action.openLargeOutput?e166cfa5-56f6-4eeb-9955-404b39db532e)---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
Cell In[37], line 10
      1 from rastervision.core.data import SemanticSegmentationLabels
      3 predictions = learner.predict_dataset(
      4     ds,
      5     raw_out=True,
      6     numpy_out=True,
      7     predict_kw=dict(out_shape=(325, 325)),
      8     progress_bar=True)
---> 10 pred_labels = SemanticSegmentationLabels.from_predictions(
     11     ds.windows,
     12     predictions,
     13     smooth=True,
     14     extent=ds.scene.extent,
     15     num_classes=len(class_config))

File [~/Desktop/BSCSFYP/RasterVision/.venv/lib/python3.8/site-packages/rastervision/core/data/label/semantic_segmentation_labels.py:220](https://file+.vscode-resource.vscode-cdn.net/home/muneeb/Desktop/BSCSFYP/RasterVision/~/Desktop/BSCSFYP/RasterVision/.venv/lib/python3.8/site-packages/rastervision/core/data/label/semantic_segmentation_labels.py:220), in SemanticSegmentationLabels.from_predictions(cls, windows, predictions, extent, num_classes, smooth, crop_sz)
    195 """Instantiate from windows and their corresponding predictions.
    196 
    197 Args:
   (...)
    217         SemanticSegmentationDiscreteLabels.
    218 """
    219 labels = cls.make_empty(extent, num_classes, smooth=smooth)
--> 220 labels.add_predictions(windows, predictions, crop_sz=crop_sz)
    221 return labels

File [~/Desktop/BSCSFYP/RasterVision/.venv/lib/python3.8/site-packages/rastervision/core/data/label/semantic_segmentation_labels.py:247](https://file+.vscode-resource.vscode-cdn.net/home/muneeb/Desktop/BSCSFYP/RasterVision/~/Desktop/BSCSFYP/RasterVision/.venv/lib/python3.8/site-packages/rastervision/core/data/label/semantic_segmentation_labels.py:247), in SemanticSegmentationLabels.add_predictions(self, windows, predictions, crop_sz)
    243     windows, predictions = discard_prediction_edges(
    244         windows, predictions, crop_sz)
    245 # If predictions is tqdm-wrapped, it needs to be the first arg to zip()
    246 # or the progress bar won't terminate with the correct count.
--> 247 for prediction, window in zip(predictions, windows):
    248     self[window] = prediction

File [~/Desktop/BSCSFYP/RasterVision/.venv/lib/python3.8/site-packages/tqdm/std.py:1195](https://file+.vscode-resource.vscode-cdn.net/home/muneeb/Desktop/BSCSFYP/RasterVision/~/Desktop/BSCSFYP/RasterVision/.venv/lib/python3.8/site-packages/tqdm/std.py:1195), in tqdm.__iter__(self)
   1192 time = self._time
   1194 try:
-> 1195     for obj in iterable:
   1196         yield obj
   1197         # Update and possibly print the progressbar.
   1198         # Note: does not call self.update(1) for speed optimisation.

File [~/Desktop/BSCSFYP/RasterVision/.venv/lib/python3.8/site-packages/rastervision/pytorch_learner/learner.py:735](https://file+.vscode-resource.vscode-cdn.net/home/muneeb/Desktop/BSCSFYP/RasterVision/~/Desktop/BSCSFYP/RasterVision/.venv/lib/python3.8/site-packages/rastervision/pytorch_learner/learner.py:735), in (.0)
    733 if numpy_out:
    734     if return_format == 'z':
--> 735         preds = (self.output_to_numpy(p) for p in preds)
    736     else:
    737         # only convert z
    738         preds = ((*p[:-1], self.output_to_numpy(p[-1])) for p in preds)

File [~/Desktop/BSCSFYP/RasterVision/.venv/lib/python3.8/site-packages/rastervision/pytorch_learner/learner.py:796](https://file+.vscode-resource.vscode-cdn.net/home/muneeb/Desktop/BSCSFYP/RasterVision/~/Desktop/BSCSFYP/RasterVision/.venv/lib/python3.8/site-packages/rastervision/pytorch_learner/learner.py:796), in (.0)
    794     preds = ((y, z) for _, y, z in preds)
    795 elif return_format == 'z':
--> 796     preds = (z for _, _, z in preds)
    798 return preds

File [~/Desktop/BSCSFYP/RasterVision/.venv/lib/python3.8/site-packages/rastervision/pytorch_learner/learner.py:831](https://file+.vscode-resource.vscode-cdn.net/home/muneeb/Desktop/BSCSFYP/RasterVision/~/Desktop/BSCSFYP/RasterVision/.venv/lib/python3.8/site-packages/rastervision/pytorch_learner/learner.py:831), in Learner._predict_dataloader(self, dl, raw_out, batched_output, predict_kw)
    829 for x, y in dl:
    830     x = self.to_device(x, self.device)
--> 831     z = self.predict(x, raw_out=raw_out, **predict_kw)
    832     x = self.to_device(x, 'cpu')
    833     y = self.to_device(y, 'cpu') if y is not None else y

File [~/Desktop/BSCSFYP/RasterVision/.venv/lib/python3.8/site-packages/rastervision/pytorch_learner/semantic_segmentation_learner.py:69](https://file+.vscode-resource.vscode-cdn.net/home/muneeb/Desktop/BSCSFYP/RasterVision/~/Desktop/BSCSFYP/RasterVision/.venv/lib/python3.8/site-packages/rastervision/pytorch_learner/semantic_segmentation_learner.py:69), in SemanticSegmentationLearner.predict(self, x, raw_out, out_shape)
     66 x = self.to_device(x, self.device)
     68 with torch.inference_mode():
---> 69     out = self.model(x)
     70     out = self.post_forward(out)
     71     out = out.softmax(dim=1)

File [~/Desktop/BSCSFYP/RasterVision/.venv/lib/python3.8/site-packages/torch/nn/modules/module.py:1130](https://file+.vscode-resource.vscode-cdn.net/home/muneeb/Desktop/BSCSFYP/RasterVision/~/Desktop/BSCSFYP/RasterVision/.venv/lib/python3.8/site-packages/torch/nn/modules/module.py:1130), in Module._call_impl(self, *input, **kwargs)
   1126 # If we don't have any hooks, we want to skip the rest of the logic in
   1127 # this function, and just call forward.
   1128 if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1129         or _global_forward_hooks or _global_forward_pre_hooks):
-> 1130     return forward_call(*input, **kwargs)
   1131 # Do not call functions when jit is used
   1132 full_backward_hooks, non_full_backward_hooks = [], []

File [~/Desktop/BSCSFYP/RasterVision/.venv/lib/python3.8/site-packages/torch/nn/modules/container.py:139](https://file+.vscode-resource.vscode-cdn.net/home/muneeb/Desktop/BSCSFYP/RasterVision/~/Desktop/BSCSFYP/RasterVision/.venv/lib/python3.8/site-packages/torch/nn/modules/container.py:139), in Sequential.forward(self, input)
    137 def forward(self, input):
    138     for module in self:
--> 139         input = module(input)
    140     return input

File [~/Desktop/BSCSFYP/RasterVision/.venv/lib/python3.8/site-packages/torch/nn/modules/module.py:1130](https://file+.vscode-resource.vscode-cdn.net/home/muneeb/Desktop/BSCSFYP/RasterVision/~/Desktop/BSCSFYP/RasterVision/.venv/lib/python3.8/site-packages/torch/nn/modules/module.py:1130), in Module._call_impl(self, *input, **kwargs)
   1126 # If we don't have any hooks, we want to skip the rest of the logic in
   1127 # this function, and just call forward.
   1128 if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1129         or _global_forward_hooks or _global_forward_pre_hooks):
-> 1130     return forward_call(*input, **kwargs)
   1131 # Do not call functions when jit is used
   1132 full_backward_hooks, non_full_backward_hooks = [], []
...
File [~/.cache/torch/hub/AdeelH_pytorch-fpn_0.3/fpn/layers.py:54](https://file+.vscode-resource.vscode-cdn.net/home/muneeb/Desktop/BSCSFYP/RasterVision/~/.cache/torch/hub/AdeelH_pytorch-fpn_0.3/fpn/layers.py:54), in Sum.forward(self, inps)
     53 def forward(self, inps):
---> 54     return sum(inps)

RuntimeError: The size of tensor a (82) must match the size of tensor b (64) at non-singleton dimension 3

[AdeelH]

The error is due to some inconsistencies between your parameter values. See the code below where I have introduced a couple of new variables to clarify which values should match.

# new variables
chip_size = <any value>
model_input_size = <any value>

model = torch.hub.load(
    ...
    out_size=(model_input_size, model_input_size))

train_ds = SemanticSegmentationRandomWindowGeoDataset.from_uris(
    ...
    out_size=model_input_size)

val_ds = SemanticSegmentationSlidingWindowGeoDataset.from_uris(
    ...
    size=chip_size,
    transform=A.Resize(model_input_size, model_input_size))

pred_ds = SemanticSegmentationSlidingWindowGeoDataset.from_uris(
    ...
    size=chip_size,
    transform=A.Resize(model_input_size, model_input_size))

predictions = learner.predict_dataset(
    ...
    predict_kw=dict(out_shape=(chip_size, chip_size)))