chenjiashuo/T2exture_model

T2exture Checkpoints

This model repo contains the released T2exture Stage 2 checkpoints:

t2exture-s.pt    T2exture-S, AMT-S backbone
t2exture-l.pt    T2exture-L, AMT-L backbone
t2exture-g.pt    T2exture-G, AMT-G backbone

Each checkpoint contains the full T2exture model state needed by infer.py and eval.py.

Expected Local Layout

Place the downloaded files under the code repository as:

pretrained/t2exture_model/t2exture-s.pt
pretrained/t2exture_model/t2exture-l.pt
pretrained/t2exture_model/t2exture-g.pt

The code does not require automatic checkpoint download. Pass the checkpoint path explicitly:

python -B infer.py \
  --mode synthetic \
  --variant l \
  --checkpoint pretrained/t2exture_model/t2exture-l.pt \
  --data-root datasets \
  --source-off-root datasets/source_off/amt-l \
  --split test \
  --output-dir outputs/infer/t2exture-l

For evaluation:

python -B eval.py \
  --data-root datasets \
  --checkpoint pretrained/t2exture_model/t2exture-l.pt \
  --backbone amt-l \
  --source-off-root datasets/source_off/amt-l \
  --split test \
  --output-dir outputs/eval/t2exture-l \
  --config train.yaml

Use the matching Stage 1 source-off cache for each variant:

t2exture-s.pt -> datasets/source_off/amt-s
t2exture-l.pt -> datasets/source_off/amt-l
t2exture-g.pt -> datasets/source_off/amt-g

Not Included

This model repo should only contain the three released T2exture checkpoints and this README. The dataset, AMT initialization checkpoints, third-party source code, logs, metrics, and qualitative outputs should stay outside the model repo.