PyTorch conversion error - TypeError with bool

[lp55]

Hi,

I'm trying to convert a model from https://github.com/ultralytics/yolov5/ (yolov5l in specific) and I get the following error:

WARNING:root:Tuple detected at graph output. This will be flattened in the converted model.
Converting Frontend ==> MIL Ops: 61%|██████████████████████████ | 1399/2307 [00:13<00:13, 64.92 ops/s]WARNING:root:Saving value type of float16 into a builtin type of i8, might lose precision!
Converting Frontend ==> MIL Ops: 61%|██████████████████████████▎ | 1414/2307 [00:13<00:13, 64.09 ops/s]WARNING:root:Saving value type of float16 into a builtin type of i8, might lose precision!
Converting Frontend ==> MIL Ops: 71%|██████████████████████████████▍ | 1636/2307 [00:16<00:11, 59.18 ops/s]WARNING:root:Saving value type of float16 into a builtin type of i8, might lose precision!
Converting Frontend ==> MIL Ops: 72%|██████████████████████████████▊ | 1651/2307 [00:17<00:11, 56.47 ops/s]WARNING:root:Saving value type of float16 into a builtin type of i8, might lose precision!
Converting Frontend ==> MIL Ops: 100%|██████████████████████████████████████████▉| 2305/2307 [00:30<00:00, 75.65 ops/s]
Running MIL optimization passes: 100%|████████████████████████████████████████████| 13/13 [00:22<00:00, 1.71s/ passes]
Translating MIL ==> MLModel Ops: 31%|████████████▏ | 553/1807 [00:00<00:00, 547815.33 ops/s]
Traceback (most recent call last):
File "", line 1, in
File "C:\Miniconda3\envs\pytorch15\lib\site-packages\coremltools\converters_converters_entry.py", line 292, in convert
proto_spec = _convert(
File "C:\Miniconda3\envs\pytorch15\lib\site-packages\coremltools\converters\mil\converter.py", line 122, in _convert
out = backend_converter(prog, **kwargs)
File "C:\Miniconda3\envs\pytorch15\lib\site-packages\coremltools\converters\mil\converter.py", line 72, in call
return load(*args, **kwargs)
File "C:\Miniconda3\envs\pytorch15\lib\site-packages\coremltools\converters\mil\backend\nn\load.py", line 235, in load
convert_ops(
File "C:\Miniconda3\envs\pytorch15\lib\site-packages\coremltools\converters\mil\backend\nn\op_mapping.py", line 50, in convert_ops
mapper(const_context, builder, op)
File "C:\Miniconda3\envs\pytorch15\lib\site-packages\coremltools\converters\mil\backend\nn\op_mapping.py", line 881, in slice_by_index
builder.add_slice_dynamic(
File "C:\Miniconda3\envs\pytorch15\lib\site-packages\coremltools\models\neural_network\builder.py", line 5501, in add_slice_dynamic
spec_layer_params.endMasks.extend(end_masks)
File "C:\Miniconda3\envs\pytorch15\lib\site-packages\google\protobuf\internal\containers.py", line 282, in extend
new_values = [self._type_checker.CheckValue(elem) for elem in elem_seq_iter]
File "C:\Miniconda3\envs\pytorch15\lib\site-packages\google\protobuf\internal\containers.py", line 282, in
new_values = [self.type_checker.CheckValue(elem) for elem in elem_seq_iter]
File "C:\Miniconda3\envs\pytorch15\lib\site-packages\google\protobuf\internal\type_checkers.py", line 142, in CheckValue
raise TypeError(message)
TypeError: True has type <class 'numpy.bool'>, but expected one of: (<class 'bool'>, <class 'numbers.Integral'>)

The code used to convert the model is the following:

from models.common import *
import coremltools as ct

img = torch.zeros((1, 3, *[640, 640]))
path = 'C:\yolov5l.pt'

model = torch.load(path, map_location=torch.device('cpu'))['model'].float()
model.eval()
model.model[-1].export = True # set Detect() layer export=True
y = model(img)
ts = torch.jit.trace(model, img)

model = ct.convert(ts, inputs=[ct.ImageType(name='images', shape=img.shape, scale=1 / 255.0, bias=[0, 0, 0])])

Seems that a simple type cast would resolve this issue. I'm not familiar with the coremltools codebase, so hopefully someone that reads this bug report knows where to fix this issue. If not could you at least point me in the right direction, so I can try to fix it myself?

[DawerG]

Thanks @lp55 for raising the issue. We are woking on the fix.

For the time being, to unblock yourself, you might want to manually change list of numpy bool to list of python boolfor end_masks object in the call spec_layer_params.endMasks.extend(end_masks)

[lp55]

@DawerG

Thanks, the conversion process worked now. I'll test the coverted model and check if it works properly now.
For any one stuck on the same issue (until the propoer fix is avaliable), I simply added the following line:

end_masks_conv = [bool(val) for val in end_masks]

and changed

spec_layer_params.endMasks.extend(end_masks)

to

spec_layer_params.endMasks.extend(end_masks_conv)

inside the add_slice_dynamic function from the builder.py file.

Not pretty but it works :)

[TobyRoseman]

Is this still an issue with the latest version of coremltools?

If it is still an issue, please give us complete steps to reproduce the problem. How was yolov5l.pt generated?