ValueError: Torch var stride_width.1 not found in context #1788
Comments
|
The Are you sure your PyTorch model is valid for the input shape you are using? |
|
Thank you @TobyRoseman for reaching out.
I'm using correct valid input shape. Here is Thats what I'm using it |
|
Right, as I said, all of the output tensors are empty. Also the fact that the first tensor is of shape Do you have a |
|
Sure, I've included example image here. Thank you @TobyRoseman. |
----> 1 import transforms as T
ModuleNotFoundError: No module named 'transforms'Where is this package coming from? I'm certainly familiar with the This same model is also causing a different error in #1790 |
Sorry I forgot to add. You need to install |
|
I have the There is no
|
|
@TobyRoseman , Okay I refined code to remove import coremltools as ct
import torch, torchvision
from torchvision.transforms import functional as F, InterpolationMode, transforms as T
import requests
from PIL import Image
import numpy as np
from typing import Dict, Tuple, Optional
# Image conversion tools:
class PILToTensor(torch.nn.Module):
def forward(
self, image: torch.Tensor, target: Optional[Dict[str, torch.Tensor]] = None
) -> Tuple[torch.Tensor, Optional[Dict[str, torch.Tensor]]]:
image = F.pil_to_tensor(image)
return image, target
class ConvertImageDtype(torch.nn.Module):
def __init__(self, dtype: torch.dtype) -> None:
super().__init__()
self.dtype = dtype
def forward(
self, image: torch.Tensor, target: Optional[Dict[str, torch.Tensor]] = None
) -> Tuple[torch.Tensor, Optional[Dict[str, torch.Tensor]]]:
image = F.convert_image_dtype(image, self.dtype)
return image, target
# Load the torchvision model
detector_model = torchvision.models.detection.fasterrcnn_mobilenet_v3_large_fpn(pretrained=True)
detector_model = detector_model.eval()
# Get a sample image
toTensor = T.PILToTensor()
toFloatTensor = T.ConvertImageDtype(torch.float)
url = 'http://images.cocodataset.org/val2017/000000039769.jpg'
example_image = Image.open(requests.get(url, stream=True).raw).convert("RGB")
example_image_np = np.array(example_image)
example_image_pt = toFloatTensor(toTensor(example_image))
example_image_pt = example_image_pt.unsqueeze(0)
# Run the sample through the model to demonstrate the model works
y = detector_model(example_image_pt)
# Make an adaptor to convert the model outputs to a tuple
class FasterRCNN_MobileNetV3_AdapterModel(torch.nn.Module):
"""This adapter is only here to unbox the first output."""
def __init__(self, model, w=2):
super().__init__()
self.model = model
def forward(self, x):
result = self.model(x)
return result[0]['boxes'], result[0]['labels'], result[0]['scores']
adapted_detector_model = FasterRCNN_MobileNetV3_AdapterModel(detector_model)
# Trace and convert the model using coremltools
model_to_trace = adapted_detector_model
with torch.inference_mode():
out = model_to_trace(example_image_pt)
traced_model = torch.jit.trace(model_to_trace, example_image_pt).eval()
detector_mlmodel = ct.convert(traced_model, inputs=[ct.ImageType(shape=example_image_pt.shape)])
detector_mlmodel.save("segmenter.mlmodel") |
TobyRoseman
added
the
triaged
|
I met similar issues when converting FasterRCNN before. This @ivyas21 Could you try to use torch 1.11.0 with torchvision 0.12.0 to see if it works? |
|
@junpeiz , Thank you. Let me try it. Hope it works as you suggest!! |
|
@junpeiz , I just downgraded torch and vision as you suggested. I'm getting this error now:
|
|
@ivyas21 Good, now we can confirm that For the Let's keep this thread focus on the |
|
@junpeiz, Sounds good. Thank you for clarifying it. I'll open new issue for |
|
@junpeiz I tried with main (e0f8918) and torch==1.13.1, I got |
When
converting a tracedtorchvisionmodel, an expected input to amuloperation is not found:ValueError: Torch var stride_width.1 not found in contextStack Trace
Steps To Reproduce
System environment:
coremltoolsversion: 6.2Linux foohostname 4.19.0-23-cloud-amd64 #1 SMP Debian 4.19.269-1 (2022-12-20) x86_64 GNU/Linux)coremltools:Please advise. Thank you!
The text was updated successfully, but these errors were encountered: