pytorch/torchvision on anaconda.org depends on ffmpeg
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 17.9k
- Forks
- 7.3k
- Avg merge
- 1d 15h
- Merged PRs (30d)
- 13
Description
🐛 Describe the bug
I'm assuming https://anaconda.org/pytorch/torchvision is where your own published packages go (as opposed to conda-forge, which is not relevant here).
Based on other discussions relating to ffmpeg, GPL, patents, etc., my understanding is that the torchvision packages should not be requiring ffmpeg to install, but will use it if it's available. This enables users with licensing concerns to simply omit it.
I also see from those discussions that conda-forge and PyPI may have different requirements. That's fine, I'm definitely 100% talking about https://anaconda.org/pytorch/torchvision only and not those other indexes.
Checking the 0.15.2 release (latest) for Linux, I see ffmpeg is listed in the index.json file (listed below). This makes it very difficult to install using conda without also installing ffmpeg - specifically, I have to install torchvision by itself and list its other dependencies manually, which is only doable with custom scripting and not a standard environment definition or metapackage.
Can we get builds that do not list ffmpeg as a dependency? It seems that was the intention, so I'm guessing something slipped in at some point.
{
"arch": "x86_64",
"build": "py38_cu118",
"build_number": 0,
"constrains": [
"cpuonly <0"
],
"depends": [
"ffmpeg >=4.2",
"jpeg",
"libpng",
"numpy >=1.11",
"pillow >=5.3.0,!=8.3.*",
"python >=3.8,<3.9.0a0",
"pytorch 2.0.1",
"pytorch-cuda 11.8.*",
"pytorch-mutex 1.0 cuda",
"requests"
],
"license": "BSD",
"name": "torchvision",
"platform": "linux",
"subdir": "linux-64",
"timestamp": 1682362621404,
"version": "0.15.2"
}
Versions
Checked with linux-64/torchvision-0.15.2-py38_cu118.tar.bz2 at https://anaconda.org/pytorch/torchvision/files
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by inspecting the referenced linux-64 torchvision 0.15.2 package and its index.json metadata, then trace the build configuration that produces its dependency list. Compare the intended optional-ffmpeg behavior with the published package metadata; done means a rebuilt package no longer requires ffmpeg while retaining the other stated dependencies.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- anaconda, python
- Domain
- build-system, machine-learning
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100