GeoSAM plugin fails to load ValueError

[klara4390]

Hi - I am quite new to using python - so I probably don't have all the necessary details included in the first try - let me know what is missing.

Trying to load the plugin from the QGIS plugin manager gives following error:
ValueError: numpy.dtype size changed, may indicate binary incompatibility. Expected 96 from C header, got 88 from PyObject

I am using 3.34.10-Prizren and Python 3.9.5.

Hope someone can help :)

[Fanchengyan]

Hi @klara4390

Did you adhere to the installation steps on https://geo-sam.readthedocs.io/en/latest/installation.html? Besides, please share the full error message to help us diagnose the problem.

[klara4390]

I did - Today QGIS is for some reason able to load the plugin, but the following error message is given when testing segmentation on the beiluhe image.

From "python error":
raceback (most recent call last):
File "/Users/klarasorensen/Library/Application Support/QGIS/QGIS3/profiles/default/python/plugins/Geo-SAM/geo_sam_tool.py", line 403, in execute_segmentation
if not self.sam_model.sam_predict(
File "/Users/klarasorensen/Library/Application Support/QGIS/QGIS3/profiles/default/python/plugins/Geo-SAM/tools/SAMTool.py", line 94, in sam_predict
self.sample = self.test_features[query]
File "/Users/klarasorensen/Library/Application Support/QGIS/QGIS3/profiles/default/python/plugins/Geo-SAM/tools/torchgeo_sam.py", line 376, in getitem
tensor = torch.tensor(dest) # .float()
RuntimeError: Could not infer dtype of numpy.float32

[klara4390]

And similar error when trying to encode an image:

Traceback (most recent call last):
File "/Users/klarasorensen/Library/Application Support/QGIS/QGIS3/profiles/default/python/plugins/Geo-SAM/tools/sam_processing_algorithm.py", line 527, in processAlgorithm
for current, batch in enumerate(ds_dataloader):
File "/Applications/QGIS-LTR.app/Contents/MacOS/lib/python3.9/site-packages/torch/utils/data/dataloader.py", line 631, in next
data = self._next_data()
File "/Applications/QGIS-LTR.app/Contents/MacOS/lib/python3.9/site-packages/torch/utils/data/dataloader.py", line 675, in _next_data
data = self._dataset_fetcher.fetch(index) # may raise StopIteration
File "/Applications/QGIS-LTR.app/Contents/MacOS/lib/python3.9/site-packages/torch/utils/data/_utils/fetch.py", line 51, in fetch
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/Applications/QGIS-LTR.app/Contents/MacOS/lib/python3.9/site-packages/torch/utils/data/_utils/fetch.py", line 51, in
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/Users/klarasorensen/Library/Application Support/QGIS/QGIS3/profiles/default/python/plugins/Geo-SAM/tools/torchgeo_sam.py", line 485, in getitem
tensor = torch.tensor(dest) # .float()
RuntimeError: Could not infer dtype of numpy.int32

Execution failed after 9.70 seconds

[Fanchengyan]

This issue appears to stem from an incompatibility between certain versions of PyTorch and NumPy.

https://stackoverflow.com/questions/74722112/runtimeerror-could-not-infer-dtype-of-numpy-float32-when-trying-to-learn-ml-a

Unity-Technologies/ml-agents#6010

Try to update or downgrade your NumPy version to see if it works.

[Fanchengyan]

Hi @klara4390

I encountered this issue on macOS as well. I was able to resolve it by updating the versions of numpy and pandas:

cd /Applications/QGIS.app/Contents/MacOS/bin

./pip3 install 'numpy<2' pandas -U