src/models/__init__.py

import os
import torch
from huggingface_hub import snapshot_download
from src.models.hmr2 import HMR2

from huggingface_hub import login
login(token=os.getenv('hf_token'))

class Namespace(dict):
    __getattr__ = dict.get


# def load_hmr2(checkpoint_path="data/model.ckpt"):
def load_hmr2(repo_id="genzoo-org/genzoo", filename="model.ckpt"):
    local_dir = snapshot_download(repo_id=repo_id)
    checkpoint_path = os.path.join(local_dir, filename)
    model_cfg = Namespace(
        MODEL=Namespace(
            BACKBONE=Namespace(
                TYPE="vit",
            ),
            BBOX_SHAPE=[192, 256],
            IMAGE_MEAN=[0.485, 0.456, 0.406],
            IMAGE_SIZE=256,
            IMAGE_STD=[0.229, 0.224, 0.225],
            SMPL_HEAD=Namespace(
                IN_CHANNELS=2048,
                TRANSFORMER_DECODER=Namespace(
                    context_dim=1280,
                    depth=6,
                    dim_head=64,
                    dropout=0.0,
                    emb_dropout=0.0,
                    heads=8,
                    mlp_dim=1024,
                    norm="layer",
                ),
                TYPE="transformer_decoder",
            ),
        ),
        SMPL=Namespace(
            NUM_BETAS=145,
            NUM_BODY_JOINTS=34,
        ),
        ckpt_path=checkpoint_path,
    )

    model = HMR2(model_cfg)
    model.load_state_dict(torch.load(checkpoint_path))
    return model, model_cfg