Add pyproject.toml for easy install

[alex-bene]

Add a pyproject.toml file to make the code easier to install for development, but also easy to install as a package.
With this, we can install locally for dev with:

uv sync --extra cu128 # or cu124, cu126, cpu depending on your cuda version

and install as a package with:

uv add git+https://github.com/facebookresearch/sam-3d-body.git

Solves: #43
Related with: #86

[victkk]

this doesn't work. it tries to install detectron before torch is installed

[alex-bene]

Hello @victkk , thanks for the comment. Could you share the command you ran to install? Indeed there was an error in the pyproject.toml for me but wasn't related to detectron but rather xtcocotools. Let me know if this fixes your issue.

[lizelive]

that patch for detectron2 was merged in

[hiAndrewQuinn]

Adding some notes from my attempts to build this repository using uv here. I was building this on the cheapest per-hour Runpod pod.

#!/bin/bash

# root assumed!
apt update && apt install -y libgl1-mesa-glx libglib2.0-0 curl git build-essential

# first get uv
curl -LsSf https://astral.sh/uv/install.sh | sh
source $HOME/.local/bin/env

# now get the latest commit of this pr
git clone https://github.com/facebookresearch/sam-3d-body
cd sam-3d-body
git fetch origin pull/88/head:pr-88
git checkout pr-88
git reset --hard 7c574e15668c0cb2eb67d714a0a7cdaed6f2fdc5

# now install sam-3d-body with this uv
uv venv

# We must manually inject the build-backend tools, because 
# --no-build-isolation **disables** the automatic provision of these tools.
uv pip install \
    hatchling \
    hatch-vcs \
    editables \
    setuptools \
    wheel \
    torch --extra-index-url https://download.pytorch.org/whl/cu124

uv sync --extra cu124 --no-build-isolation

This took a long time, but it did work eventually.