Hybrid pipeline (CV + CLIP + YOLO) class folder

This folder is separate from the original FRAMED repo code. It just runs a small 3-step pipeline and prints/saves a JSON report.

What it does

• Path A (classical CV): gets deterministic metrics from OpenCV
  • green coverage % (HSV mask)
  • laplacian variance (texture proxy)
  • canny edge density (structure proxy)
• Path B (semantic): two small pretrained models
  • CLIP zero-shot scene tag (hardcoded label list)
  • YOLOv8n detections (objects + boxes)
• Fusion: hardcoded rules (if/else) combine it into final report + notes

Install

From repo root:

pip install -r moiz_kashif_csc481_final/requirements.txt

Run

Run 1 image:

python moiz_kashif_csc481_final/pipeline.py --image "path/to/img.jpg" --out_dir moiz_kashif_csc481_final/sample_outputs

Run a folder:

python moiz_kashif_csc481_final/pipeline.py --input_dir moiz_kashif_csc481_final/dataset/success_cases --out_dir moiz_kashif_csc481_final/sample_outputs

Extra outputs for presentation

For each image, the pipeline writes:

• *.json – full report (Path A metrics, CLIP, YOLO, quality_grade, timing)
• *.slim.json – compact slide-friendly summary
• *_green_mask.png – HSV green mask (vegetation pixels)
• *_edges.png – Canny edge map (structure)
• *_hud.jpg – visual HUD overlay (YOLO boxes, CLIP scene, LapVar/edges, PASS/REVIEW/REJECT)

When running on a folder (--input_dir), a simple CSV is also written:

• master_culling_report.csv in the out_dir, with columns:
  • filename, quality_grade, scene, laplacian_variance, edge_density_tuned, primary_subject

Auto-tests (HW-style rubric)

Run the deterministic CV-only tests (fast, no model downloads):

pytest -q moiz_kashif_csc481_final/tests

Run the optional CLIP+YOLO integration test (may download weights/models on first run):

RUN_SEMANTICS_TESTS=1 pytest -q moiz_kashif_csc481_final/tests/test_pipeline_schema_optional.py

Notes

• YOLO uses yolov8n.pt and will auto-download it on first run.
• CLIP labels are fixed inside semantics_utils.py (so its not open ended).