README command validity checker found 2 blocking issues (env name mismatch, missing training entrypoint)

[bbbbinzhao]

Hi maintainers,

While following the README to set up and train DiffusionTrack, I found two issues that make the provided commands fail out-of-the-box. I detected them using a lightweight automated “README command validity checker” that parses shell commands in README and performs basic consistency checks (e.g., conda env naming consistency, referenced script/path existence).

---

Issue 1: Conda environment name mismatch (create -n vs activate)

Where

README environment setup section shows:

conda create -n seqtrack python=3.8
conda activate diffusintrack

Problem

The environment name created (seqtrack) does not match the name being activated (diffusintrack). Following the README literally causes activation to fail.

Impact

Users following the instructions cannot proceed past environment activation.

Suggested fix

Make the environment name consistent, e.g. change:

conda activate diffusintrack → conda activate seqtrack

(or rename the conda create -n ... line to match, as long as both are identical).

---

Issue 2: Training command references a non-existent script lib/train/run_training_diffusiontrack.py

Where

README “Train DiffusionTrack” provides a command similar to:

python -m torch.distributed.launch --nproc_per_node 8 \
  lib/train/run_training_diffusiontrack.py \
  --script diffusiontrack --config diffusiontrack_b256 --save_dir ./output/diffusiontrack_b256

Problem

In the repository checkout, lib/train/run_training_diffusiontrack.py does not exist. Under lib/train/ I only see run_training.py (no run_training_diffusiontrack.py).

Impact

The training command fails immediately because Python cannot open the referenced file.

Suggested fix (either option is fine)

Update README to point to the correct existing entry script (if lib/train/run_training.py is the intended one), e.g.:

python -m torch.distributed.launch --nproc_per_node 8 \
  lib/train/run_training.py \
  --script diffusiontrack --config diffusiontrack_b256 --save_dir ./output/diffusiontrack_b256

Or, add lib/train/run_training_diffusiontrack.py as a thin wrapper that delegates to the real training entrypoint (useful for backward compatibility with existing docs/scripts).

---

Automated checker summary (for reference)

ERROR (R1 CondaEnvNameMismatch): created env seqtrack but activated diffusintrack

ERROR (R2 ReferencedScriptNotFound): referenced script lib/train/run_training_diffusiontrack.py not found in repo

Thanks for the great project!