[harshi-puli]

Summary

• Organized gate YOLO files under perception/tasks/gate/yolo/ so source code, configs, tests, local datasets, and model weights have clearer homes.
• Moved dataset configs into perception/tasks/gate/yolo/configs/.
• Moved local sample data into the repo-level ignored data/ directory.
• Moved local model weights into data/gate_yolo_weights/.
• Refactored split_gate_data.py into an import-safe, testable dataset splitter with CLI arguments.
• Added gate YOLO tests, including splitter tests and a guarded YOLO prediction smoke test.
• Made optional saliency/archive imports lazy in perception/__init__.py so tests can collect without optional compiled extensions.
• Added repo hygiene READMEs for:
  • perception/misc/
  • perception/third_party/

The new READMEs document what belongs in each folder, what currently lives there, and how future integration should be handled.

Testing

pytest perception/tasks/gate/tests/test_yolo_split.py \
       perception/tasks/gate/tests/test_yolo_predict_smoke.py
## How to test

```bash
cd "perception"

# Run the gate YOLO splitter tests and prediction smoke test
pytest perception/tasks/gate/tests/test_yolo_split.py \
       perception/tasks/gate/tests/test_yolo_predict_smoke.py

Expected result without YOLO dependencies installed:

2 passed, 1 skipped

The skipped test is the YOLO inference smoke test. It skips automatically when ultralytics is not installed.

How to test YOLO prediction locally

Install the YOLO/Torch dependencies first:

uv pip install -r requirements-torch.txt

Or, if not using uv:

pip install ultralytics

Then run the smoke test again:

pytest perception/tasks/gate/tests/test_yolo_predict_smoke.py

The test loads:

data/gate_yolo_weights/yolo11n.pt

and predicts on two images from:

data/gate_yolo_sample/test/images/

Manual prediction command

python perception/tasks/gate/yolo/predict_gate.py \
  --source data/gate_yolo_sample/test/images \
  --model data/gate_yolo_weights/yolo11n.pt \
  --output data/gate_yolo_predictions

Predictions are written to:

data/gate_yolo_predictions/run/

Dataset splitting command

To split a generated YOLO-style dataset with images/ and labels/ folders:

python -m perception.tasks.gate.yolo.split_gate_data \
  --source path/to/generated/gate_dataset \
  --output data/gate_yolo_sample

By default, this creates:

train/images
train/labels
val/images
val/labels
test/images
test/labels

Notes

• The local sample dataset and YOLO weights live under data/, which is already gitignored.
• The prediction smoke test is intentionally guarded so CI or local classical-only environments do not fail when YOLO dependencies are missing.
• Torch/YOLO CI coverage is not added in this PR.

[kvaikunthan]

ts tuff af