facebookresearch / facebookresearch/detectron2
PEP 503
- Dominant language
- Python
- Stars
- 34.7k
- Forks
- 7.9k
- PR merge metrics
- No merged PRs in 30d
Description
## 🚀 Feature
Provide pre-built detectron2 in [PEP 503](https://www.python.org/dev/peps/pep-0503/) simple repository.
## Motivation & Examples
Provide PEP 503 compliant simple repository for pre-built detectron2, allowing use of the pip `--extra-index-url` argument or the a pipenv `[[source]]` configuration. This allows pip/pipenv to automatically select the appropriate prebuilt binary for the user's python version without exact manual specification of the full binary file name.
For example, to install detectron v0.6 for cuda 11.3 and torch 1.10 via pip:
```
pip install --extra-index-url https://dl.fbaipublicfiles.com/detectron2/wheels/cu113/torch1.10/ detectron2==0.6+cu113
```
and via pipenv
```
[[source]]
name = "detectron2"
url = "https://dl.fbaipublicfiles.com/detectron2/wheels/cu113/torch1.10/"
verify_ssl = true
[packages]
torch = {version="0.6+cu113", index="detectron2"}
```
Related links regarding PEP 503 implementation for pytorch:
[PEP 503 for pytorch](https://github.com/pytorch/pytorch/issues/25639)
[Installing torch with pipenv](https://github.com/pypa/pipenv/issues/4961#issuecomment-1045679643)
## Extra Credit
Reorganize/rename prebuilt detectron2 for a simplified index url and more descriptive package version. All prebuilt packages could then be hosted under a single url (`https://dl.fbaipublicfiles.com/wheels/detectron2/`).
```
pip install --extra-index-url https://dl.fbaipublicfiles.com/wheels/ detectron2==0.6+cu113+torch110
```
Contributor guide
Research direction
The issue names no repository files or tests. Start by reviewing how detectron2 wheels are currently hosted and organized, then read PEP 503 and the linked PyTorch implementation issue. Done means a compliant simple repository lets pip and pipenv select the appropriate pre-built detectron2 package through an extra index or source URL.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, pytorch
- Domain
- build-system, machine-learning
- Issue type
- Feature
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100