#!/opt/homebrew/bin/python3.11
"""
Test Grounding DINO Large with COMBINED prompts — one inference per frame.
"""
import json, os, time, cv2, torch
from PIL import Image
from transformers import AutoProcessor, AutoModelForZeroShotObjectDetection

MODEL_PATH = "/Users/accusys/momentry_core_0.1/models/gun/grounding-dino-large-hf"
VIDEO = "/Users/accusys/momentry/var/sftpgo/data/demo/Charade (1963) Cary Grant & Audrey Hepburn \uff5c Comedy Mystery Romance Thriller \uff5c Full Movie.mp4"
OUTPUT_DIR = "/Users/accusys/momentry/output_dev/zero_shot_objects"
os.makedirs(OUTPUT_DIR, exist_ok=True)

TIMEPOINTS = [
    (429, "stamp"), (691, "stamp_letter"), (762, "passport"),
    (3491, "passport"), (5054, "passport"),
    (5434, "letter"), (5443, "stamp_envelope"),
    (5467, "envelope"), (5500, "stamp"), (5506, "stamp"),
    (5783, "letter"), (5786, "envelope"),
]

COMBINED_PROMPT = "stamp. postage stamp. envelope. passport. identification. letter."

print("Loading Large model...")
t0 = time.time()
processor = AutoProcessor.from_pretrained(MODEL_PATH)
model = AutoModelForZeroShotObjectDetection.from_pretrained(MODEL_PATH)
device = "mps" if torch.backends.mps.is_available() else "cpu"
model.to(device)
print(f"Loaded in {time.time()-t0:.1f}s")

cap = cv2.VideoCapture(VIDEO)
fps = cap.get(cv2.CAP_PROP_FPS) or 25.0

print(f"\nTesting {len(TIMEPOINTS)} timepoints with combined prompt...")
t_infer = time.time()

for t_sec, label in TIMEPOINTS:
    cap.set(cv2.CAP_PROP_POS_FRAMES, int(t_sec * fps))
    ret, frame = cap.read()
    if frame is None: continue

    img = Image.fromarray(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))

    # ONE inference with ALL prompts
    inputs = processor(images=img, text=COMBINED_PROMPT, return_tensors="pt").to(device)
    with torch.no_grad():
        outputs = model(**inputs)
    target = torch.tensor([img.size[::-1]])
    dets = processor.post_process_grounded_object_detection(
        outputs, threshold=0.1, target_sizes=target
    )[0]

    det_list = []
    for i in range(len(dets["boxes"])):
        det_list.append({
            "bbox": [round(v, 1) for v in dets["boxes"][i].tolist()],
            "score": round(dets["scores"][i].item(), 3),
            "label": str(dets["labels"][i]) if "labels" in dets else "object",
        })

    # Classify which expected objects were found
    found = set()
    for d in det_list:
        lbl = d["label"].lower()
        for obj in ["stamp", "envelope", "passport", "letter"]:
            if obj in lbl:
                found.add(obj)

    found_str = ", ".join(sorted(found)) if found else "none"
    print(f"  {t_sec//60}:{t_sec%60:02d} {label:20s} | {len(det_list)} dets | found: [{found_str}]")

    # Save annotated frame
    for d in det_list:
        x1, y1, x2, y2 = [int(v) for v in d["bbox"]]
        cv2.rectangle(frame, (x1, y1), (x2, y2), (0, 255, 0), 2)
        cv2.putText(frame, f"{d['label']} {d['score']:.2f}", (x1, y1-5),
                    cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)

    cv2.imwrite(os.path.join(OUTPUT_DIR, f"combined_{t_sec}s.jpg"), frame, [cv2.IMWRITE_JPEG_QUALITY, 85])

cap.release()
print(f"\nDone in {time.time()-t_infer:.0f}s")
print(f"Screenshots: {OUTPUT_DIR}/")