| """End-to-end comparison on the user's actual schematic: VLM off vs on. |
| |
| Saves two annotated PNGs so the FP reduction is directly visible. |
| """ |
| import os |
| import sys |
| import time |
|
|
| import cv2 |
|
|
| sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__)))) |
|
|
| from src.pipeline import PatternDetectionPipeline |
|
|
| DRAWING = os.environ.get("DRAWING", r"D:\Sotatek_Assessment\drawings\1.png") |
| PATTERN = r"D:\Sotatek_Assessment\drawings\test_2.png" |
| OUT = os.path.join(os.path.dirname(os.path.dirname(os.path.abspath(__file__))), |
| "debug_output") |
|
|
|
|
| def run(use_vlm: bool, tag: str): |
| print(f"\n########## use_vlm={use_vlm} ##########") |
| pipe = PatternDetectionPipeline(config={ |
| "use_vlm": use_vlm, |
| "vlm_symbol_name": "a resistor (zigzag or plain rectangle)", |
| }) |
| t0 = time.time() |
| result = pipe.detect_auto(PATTERN, DRAWING, return_visualization=True) |
| dt = time.time() - t0 |
| n = result["total_detections"] |
| outpath = os.path.join(OUT, f"compare_{tag}.png") |
| cv2.imwrite(outpath, result["visualization"]) |
| print(f"[CMP] {tag}: {n} detections in {dt:.1f}s -> {outpath}") |
| |
| classes = {} |
| for d in result["detections"]: |
| cl = d.get("vlm_class") |
| if cl: |
| classes[cl] = classes.get(cl, 0) + 1 |
| if classes: |
| print(f"[CMP] {tag} vlm_class kept: {classes}") |
| return n |
|
|
|
|
| def main(): |
| if not os.path.exists(DRAWING): |
| print(f"MISSING drawing: {DRAWING}") |
| return |
| n_off = run(False, "vlm_off") |
| n_on = run(True, "vlm_on") |
| print(f"\n========== RESULT ==========") |
| print(f" VLM off: {n_off} detections") |
| print(f" VLM on : {n_on} detections") |
| print(f" removed by VLM: {n_off - n_on}") |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
