kevin/FishServer
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

cli to control zed camera start and stop. 2. measure now use every svo2 file for 1 fish, give intermideate result and final result with confidecne level(*).

Browse Source
This commit is contained in:
kevin
2026-04-16 11:38:30 +08:00
parent 9dce487c79
commit cc6cef0f73
57 changed files with 1877 additions and 386 deletions
Download Patch File Download Diff File Expand all files Collapse all files

72
FishMeasure/generate_video_with_labels.py
View File

@@ -167,22 +167,41 @@ def generate_video(
imgsz: int = 640,
frame_stride: int = 1,
show_large: bool = False,
summary_weight_g: Optional[float] = None,
summary_length_mm: Optional[float] = None,
summary_star: bool = False,
output_video_name: Optional[str] = None,
sam_device: str = "cuda",
) -> Optional[Path]:
if not ZED_AVAILABLE:
print("ERROR: pyzed not available, cannot generate labeled video")
return None
per_frame, summary_wg, summary_lmm, is_confident = _parse_weight_json(weight_json)
star_s = " *" if is_confident else ""
per_frame, parsed_summary_wg, parsed_summary_lmm, raw_confident = _parse_weight_json(weight_json)
if summary_weight_g is None:
summary_weight_g = parsed_summary_wg
if summary_length_mm is None:
summary_length_mm = parsed_summary_lmm
star_s = " *" if summary_star else ""
print(f" Per-frame predictions: {len(per_frame)} PLYs mapped")
print(f" Summary: weight={summary_wg}g, length={summary_lmm}mm{star_s}")
print(
f" Summary: weight={summary_weight_g}g, length={summary_length_mm}mm{star_s} "
f"(raw_confident={raw_confident})"
)
if not per_frame and summary_wg is None:
if not per_frame and summary_weight_g is None:
print(" WARNING: No weight data in JSON, video will show '--'")
from ultralytics import YOLO
yolo = YOLO(yolo_model_path)
class_names = yolo.names if hasattr(yolo, "names") else {}
from fish_video_weight_evaluation import (
create_segmentation_overlay,
load_sam_predictor_with_fallback,
segment_with_sam,
)
sam_predictor, eff_sam_device = load_sam_predictor_with_fallback(sam_device)
sam_torch_device = eff_sam_device
from dataset.zed_reader import ZEDReader
zed_reader = ZEDReader(svo_path=str(svo_path), camera_mode=False, use_yolo_detector=False)
@@ -218,6 +237,7 @@ def generate_video(
continue
frame_number = idx + 1
frame_name = f"frame_{frame_number:06d}"
if frame_number in per_frame:
cur_wg, cur_lmm = per_frame[frame_number]
last_wg = cur_wg
@@ -230,6 +250,7 @@ def generate_video(
num_dets = len(results.boxes) if results.boxes is not None else 0
left_disp = img.copy()
right_disp = img.copy()
if num_dets > 0:
boxes = results.boxes.xyxy.cpu().numpy()
tids = (results.boxes.id.cpu().numpy().astype(int)
@@ -245,19 +266,31 @@ def generate_video(
cname = class_names.get(cid, "fish")
_draw_label_on_box(left_disp, box, tid, cname, cur_wg, cur_lmm)
if show_large or summary_wg is not None:
_draw_large_summary(left_disp, summary_wg, summary_lmm, is_confident)
try:
masks = segment_with_sam(sam_predictor, img, boxes, sam_torch_device)
except Exception as e:
print(f" WARNING: SAM segmentation failed on {frame_name}: {e}")
masks = []
if masks:
right_disp = create_segmentation_overlay(img.copy(), masks)
cv2.putText(right_disp, "Segmentation", (10, right_disp.shape[0] - 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255, 255, 255), 2, cv2.LINE_AA)
else:
cv2.putText(right_disp, "Segmentation (failed)", (10, right_disp.shape[0] - 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2, cv2.LINE_AA)
else:
cv2.putText(right_disp, "No detections", (10, right_disp.shape[0] - 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (128, 128, 128), 2, cv2.LINE_AA)
if show_large or summary_weight_g is not None:
_draw_large_summary(left_disp, summary_weight_g, summary_length_mm, summary_star)
frame_name = f"frame_{frame_number:06d}"
info = f"[{frame_number}] {frame_name} | Detections: {num_dets}"
cv2.putText(left_disp, info, (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 255, 0), 2, cv2.LINE_AA)
cv2.putText(left_disp, "Detection", (10, left_disp.shape[0] - 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255, 255, 255), 2, cv2.LINE_AA)
right_disp = img.copy()
cv2.putText(right_disp, "Original", (10, right_disp.shape[0] - 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255, 255, 255), 2, cv2.LINE_AA)
combined = np.hstack([left_disp, right_disp])
if num_dets > 0:
frames.append(combined)
@@ -275,7 +308,7 @@ def generate_video(
print(f" WARNING: No detection frames collected from {svo_name}")
return None
video_path = images_dir / f"{svo_name}_preview.mp4"
video_path = images_dir / (output_video_name or f"{svo_name}_preview.mp4")
h, w = frames[0].shape[:2]
writer = cv2.VideoWriter(str(video_path), cv2.VideoWriter_fourcc(*"mp4v"), 10.0, (w, h))
for f in frames:
@@ -295,7 +328,17 @@ def main():
parser.add_argument("--conf", type=float, default=0.25)
parser.add_argument("--imgsz", type=int, default=640)
parser.add_argument("--frame-stride", type=int, default=1)
parser.add_argument("--sam-device", type=str, default="cuda")
parser.add_argument("--show-large-labels-at-top-right", action="store_true")
parser.add_argument(
"--summary-star",
action=argparse.BooleanOptionalAction,
default=False,
help="Whether to draw * on the Final summary line; caller/DB is the source of truth.",
)
parser.add_argument("--summary-weight-g", type=float, default=None)
parser.add_argument("--summary-length-mm", type=float, default=None)
parser.add_argument("--output-video-name", type=str, default=None)
args = parser.parse_args()
svo = Path(args.svo).expanduser().resolve()
@@ -315,7 +358,12 @@ def main():
conf=args.conf,
imgsz=args.imgsz,
frame_stride=args.frame_stride,
sam_device=args.sam_device,
show_large=args.show_large_labels_at_top_right,
summary_weight_g=args.summary_weight_g,
summary_length_mm=args.summary_length_mm,
summary_star=bool(args.summary_star),
output_video_name=args.output_video_name,
)