mewc-torch Docker Image

This repository contains the Dockerfile and docker-compose.yml files used to build the mewc-torch Docker image. The mewc-torch Docker image serves as the base image for the mewc-detect Docker image.

• mewc-detect Docker Image Repository

The mewc-torch image is built on top of the pytorch/pytorch:2.0.1-cuda11.7-cudnn8-runtime image and includes additional dependencies required by MegaDetector v5 and YOLOv5.

MDv5/MDv1000 PyTorch-only Tag

For MDv5/MDv1000 we provide a PyTorch-only base image with explicit pins and no TensorFlow/tensorboard to avoid NumPy 2.x ABI conflicts and import-time stalls.

• Core pins:
  • NumPy: 1.26.4
  • OpenCV (headless): 4.8.1.78
  • PyTorch: 2.0.1 (CUDA 11.7)
  • TorchVision: 0.15.2
• Ultralytics: 8.0.123 (installed with --no-deps)
• No TensorFlow/tensorboard in this image.

Build locally with an explicit tag and run a quick import sanity check:

docker build -t zaandahl/mewc-torch:py310-cu117-torch2.0.1-no-tf .
docker run --rm -it zaandahl/mewc-torch:py310-cu117-torch2.0.1-no-tf python - <<'PY'
import pkgutil, numpy, torch, torchvision, cv2
print('numpy', numpy.__version__)
print('torch', torch.__version__, 'cuda', torch.version.cuda)
print('vision', torchvision.__version__)
print('cv2', cv2.__version__)
print('tensorflow_present', bool(pkgutil.find_loader("tensorflow")))
PY

MegaDetector v5

MegaDetector v5 is a version of MegaDetector that utilizes the YOLOv5 neural network for object detection. More information about MegaDetector v5 can be found in its GitHub repositories:

• MegaDetector v5 GitHub Repository

YOLOv5

YOLOv5 is a state-of-the-art, extremely fast and accurate object detection model. More information about YOLOv5 can be found in its GitHub repository:

• YOLOv5 GitHub Repository

Building the Docker Image

This repository uses docker-compose to build the Docker image. To build the mewc-torch Docker image, clone this repository and use the docker-compose command:

git clone https://github.com/zaandahl/mewc-torch.git
cd mewc-torch
docker-compose up --build

This will create a Docker image named zaandahl/mewc-torch.

Docker Image Contents

The Docker image contains:

• Python environment isolated in a venv with pinned core packages for MDv5/MDv1000 (see above).
• Necessary utilities (ffmpeg, libsm6, libxext6, git, wget) installed.
• Source code copied into the /code/src directory in the container.

For detailed information about the image contents, please refer to the Dockerfile in this repository.

GitHub Actions and DockerHub

This project uses GitHub Actions to automate the build process and push the Docker image to DockerHub. You can find the image at:

• zaandahl/mewc-torch DockerHub Repository

---

As always, please modify the README.md as necessary to better fit your project's needs.