PyTorchCV

0.0.74 verified Fri May 01 auth: no python

PyTorchCV provides a collection of computer vision models (classification, segmentation, detection, etc.) for PyTorch, with pretrained weights. Current version 0.0.74, release cadence is irregular. Requires Python >=3.10.

pip install pytorchcv

Common errors

error ModuleNotFoundError: No module named 'pytorchcv' ↓

cause Package not installed or installed in wrong environment.

fix

Run: pip install pytorchcv

error KeyError: 'resnet18' ↓

cause Using incorrect model name (case or spelling).

fix

Check list of models: from pytorchcv.model_provider import get_model_list; print(get_model_list())

error AttributeError: 'NoneType' object has no attribute 'shape' ↓

cause Model returned None because pretrained weights failed to load (network issue).

fix

Ensure internet connection or set pretrained=False to load architecture without weights.

Warnings

gotcha Model names are case-sensitive and must match exactly the keys in model_provider (e.g., 'resnet18' works, 'ResNet18' does not). ↓

fix Use lowercase names; check available models via pytorchcv.model_provider.get_model_list().

deprecated Some older model APIs (e.g., direct import of model classes from pytorchcv.models) may be deprecated in future releases. Use get_model() for future compatibility. ↓

fix Replace direct class imports with get_model('model_name').

gotcha Pretrained weights download from a remote server; if the server is down or slow, model loading may fail or hang. There is no fallback or mirror. ↓

fix Set environment variable PYTORCHCV_CACHE_DIR to customize cache location; consider pre-downloading weights.

Imports

get_model
```
from pytorchcv.model_provider import get_model
```
get_model is the primary function to load any model by name.

Quickstart

Quickstart: load a pretrained ResNet-18 model and run inference on a random tensor.

import torch
from pytorchcv.model_provider import get_model

model = get_model('resnet18', pretrained=True)
model.eval()
x = torch.randn(1, 3, 224, 224)
with torch.no_grad():
    out = model(x)
print(out.shape)