File size: 13,453 Bytes
b27cd24 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 | # batch_line_visibility.py
import os
import csv
import json
import numpy as np
import cv2
from eval_gpt import load_all_json_recursive_with_paths # same as your attached script
from identify_queue_start_end import identify_start_end_bboxes, load_fpx_from_txt
GT_FOLDER = "/scratch/ds5725/linefinder/LineFinder/GT_json"
IMG_FOLDER = "/scratch/ds5725/linefinder/LineFinder/Images"
DEPTH_DIR = "/scratch/ds5725/linefinder/LineFinder/depth_map"
BBOX_ORIENT_DIR = "/scratch/ds5725/linefinder/LineFinder/bbox_orient"
FOCAL_TXT = "/scratch/ds5725/linefinder/LineFinder/focal_length_px.txt"
def flatten_and_fix_gt(gt_entry: dict) -> dict:
"""
Input GT format (nested):
gt_entry["end_of_line"]["visible"]
gt_entry["end_of_line"]["location_if_visible"]
gt_entry["end_of_line"]["direction_to_turn_if_not_visible"]
Output (flat, matching prediction keys):
end_of_line_visible
end_of_line_location_if_visible
direction_to_turn_to_see_end_if_not_visible
Same for start.
Also enforces:
visible=="yes" => turn="N/A"
visible=="no" => location="N/A"
"""
def get_nested(side: str, key: str, default=""):
obj = gt_entry.get(f"{side}_of_line", {})
if not isinstance(obj, dict):
return default
return obj.get(key, default)
flat = {
"end_of_line_visible": str(get_nested("end", "visible", "")).strip().lower(),
"end_of_line_location_if_visible": str(get_nested("end", "location_if_visible", "N/A")).strip().lower(),
"direction_to_turn_to_see_end_if_not_visible": str(get_nested("end", "direction_to_turn_if_not_visible", "N/A")).strip().lower(),
"start_of_line_visible": str(get_nested("start", "visible", "")).strip().lower(),
"start_of_line_location_if_visible": str(get_nested("start", "location_if_visible", "N/A")).strip().lower(),
"direction_to_turn_to_see_start_if_not_visible": str(get_nested("start", "direction_to_turn_if_not_visible", "N/A")).strip().lower(),
}
# Canonicalize / repair consistency
def fix(prefix: str):
vis_k = f"{prefix}_of_line_visible"
loc_k = f"{prefix}_of_line_location_if_visible"
turn_k = f"direction_to_turn_to_see_{prefix}_if_not_visible"
vis = flat.get(vis_k, "")
if vis not in ("yes", "no"):
return
if vis == "yes":
flat[turn_k] = "N/A"
valid_locs = {"far left","center left","center","center right","far right"}
if flat.get(loc_k, "N/A") not in valid_locs:
flat[loc_k] = "N/A"
else: # vis == "no"
flat[loc_k] = "N/A"
valid_turns = {"left","right"}
if flat.get(turn_k, "N/A") not in valid_turns:
flat[turn_k] = "N/A"
fix("end")
fix("start")
# Store canonical casing (match your prediction strings)
# visible: yes/no already lowercase
# location: lowercase; N/A uppercase
for k in ["end_of_line_location_if_visible", "start_of_line_location_if_visible",
"direction_to_turn_to_see_end_if_not_visible", "direction_to_turn_to_see_start_if_not_visible"]:
v = flat.get(k, "N/A")
flat[k] = "N/A" if v in ("n/a", "na", "") else v
for k in ["end_of_line_visible", "start_of_line_visible"]:
v = flat.get(k, "")
flat[k] = "yes" if v == "yes" else ("no" if v == "no" else "")
return flat
def normalize_visibility_fields(gt: dict) -> dict:
"""
Fix inconsistent GT fields in-place according to the visibility rules.
Returns a new dict (copy) with repaired fields.
"""
gt = dict(gt) # shallow copy
def norm_side(prefix: str):
# prefix in {"start", "end"}
vis_k = f"{prefix}_of_line_visible"
loc_k = f"{prefix}_of_line_location_if_visible"
turn_k = f"direction_to_turn_to_see_{prefix}_if_not_visible"
vis = str(gt.get(vis_k, "")).strip().lower()
if vis not in ("yes", "no"):
return # leave as-is if missing/invalid
# normalize to canonical case
gt[vis_k] = "yes" if vis == "yes" else "no"
if gt[vis_k] == "yes":
# visible => turn must be N/A
gt[turn_k] = "N/A"
# location can stay if valid, otherwise N/A
valid_locs = {"far left", "center left", "center", "center right", "far right"}
loc = str(gt.get(loc_k, "N/A")).strip().lower()
if loc in valid_locs:
# store canonical case exactly
gt[loc_k] = loc
else:
gt[loc_k] = "N/A"
else:
# not visible => location must be N/A
gt[loc_k] = "N/A"
# turn can stay if valid; otherwise N/A
valid_turn = {"left", "right"}
turn = str(gt.get(turn_k, "N/A")).strip().lower()
if turn in valid_turn:
gt[turn_k] = turn
else:
gt[turn_k] = "N/A"
norm_side("end")
norm_side("start")
return gt
def get_images_with_gt(img_folder, gt_keys):
"""Same logic as in batch_queue_direction.py: match by basename (no extension)."""
matched = []
valid_exts = (".jpg", ".jpeg", ".png", ".webp", ".gif")
for root, _, files in os.walk(img_folder):
for fname in files:
if fname.lower().endswith(valid_exts):
key = os.path.splitext(fname)[0]
if key in gt_keys:
matched.append(os.path.join(root, fname))
return matched
def _bbox_edge_flags(bbox_xyxy, W, H, margin_px):
x1, y1, x2, y2 = [float(v) for v in bbox_xyxy.tolist()]
near_left = x1 <= margin_px
near_right = x2 >= (W - 1 - margin_px)
near_top = y1 <= margin_px
near_bottom = y2 >= (H - 1 - margin_px)
touches_any = near_left or near_right or near_top or near_bottom
return touches_any, near_left, near_right, near_top, near_bottom
def _location_bucket_from_center_x(cx, W):
r = cx / max(W, 1)
if r < 0.2:
return "far left"
elif r < 0.4:
return "center left"
elif r < 0.6:
return "center"
elif r < 0.8:
return "center right"
else:
return "far right"
def endpoint_fields(bbox_xyxy, W, H, margin_px):
"""
Implements your rule:
- if bbox touches/is close to any edge -> not visible
- if not visible: turn left if near left edge else right
- if visible: location bucket by bbox center x
"""
x1, y1, x2, y2 = [float(v) for v in bbox_xyxy.tolist()]
cx = 0.5 * (x1 + x2)
touches_any, near_left, near_right, near_top, near_bottom = _bbox_edge_flags(
bbox_xyxy, W, H, margin_px
)
if touches_any:
visible = "no"
location = "N/A"
turn = "left" if near_left else "right"
else:
visible = "yes"
location = _location_bucket_from_center_x(cx, W)
turn = "N/A"
return visible, location, turn
def process_one_image(img_path, gt_entry, margin_px=10):
image_id = os.path.splitext(os.path.basename(img_path))[0]
# Paths
depth_path = os.path.join(DEPTH_DIR, image_id + ".npy")
bbox_path = os.path.join(BBOX_ORIENT_DIR, image_id + "_bboxes.npy")
orient_path = os.path.join(BBOX_ORIENT_DIR, image_id + "_orient.npy")
# Required files check
for p in [depth_path, bbox_path, orient_path, FOCAL_TXT]:
if not os.path.isfile(p):
return None, f"missing:{p}"
# Read image for W,H
img = cv2.imread(img_path)
if img is None:
return None, "missing-image"
H, W = img.shape[:2]
# Look up focal length
try:
f_px = load_fpx_from_txt(FOCAL_TXT, image_id)
except Exception as e:
return None, f"missing-fpx:{e}"
# Identify start/end bboxes
try:
res = identify_start_end_bboxes(
image_path=img_path,
depth_npy_path=depth_path,
bboxes_npy_path=bbox_path,
orient_npy_path=orient_path,
f_px=f_px,
)
except Exception as e:
return None, f"fail-identify:{e}"
start_bbox = res["start_bbox_xyxy"] # START = head
end_bbox = res["end_bbox_xyxy"] # END = tail
# Compute fields
end_visible, end_loc, end_turn = endpoint_fields(end_bbox, W, H, margin_px)
start_visible, start_loc, start_turn = endpoint_fields(start_bbox, W, H, margin_px)
pred = {
"image_id": image_id,
"image_path": img_path,
"end_of_line_visible": end_visible,
"end_of_line_location_if_visible": end_loc,
"direction_to_turn_to_see_end_if_not_visible": end_turn,
"start_of_line_visible": start_visible,
"start_of_line_location_if_visible": start_loc,
"direction_to_turn_to_see_start_if_not_visible": start_turn,
}
# Pull GT fields if present
gt = {}
if isinstance(gt_entry, dict):
for k in [
"end_of_line_visible",
"end_of_line_location_if_visible",
"direction_to_turn_to_see_end_if_not_visible",
"start_of_line_visible",
"start_of_line_location_if_visible",
"direction_to_turn_to_see_start_if_not_visible",
]:
if k in gt_entry:
gt[k] = gt_entry[k]
return (pred, gt), "ok"
def main():
# Load GT jsons (same as batch_queue_direction.py)
gt_dict, gt_paths = load_all_json_recursive_with_paths(GT_FOLDER)
gt_keys = set(gt_dict.keys())
print(f"Loaded {len(gt_keys)} GT JSONs.")
# Find matching images (same logic)
image_paths = get_images_with_gt(IMG_FOLDER, gt_keys)
print(f"Found {len(image_paths)} images that have GT JSONs.")
margin_px = 10 # tweak if needed
# Collect results
rows = []
correct = {k: 0 for k in [
"end_of_line_visible",
"end_of_line_location_if_visible",
"direction_to_turn_to_see_end_if_not_visible",
"start_of_line_visible",
"start_of_line_location_if_visible",
"direction_to_turn_to_see_start_if_not_visible",
]}
total = {k: 0 for k in correct.keys()}
failures = 0
for img_path in image_paths:
image_id = os.path.splitext(os.path.basename(img_path))[0]
# 1) Build FLAT + FIXED GT (nested -> flat, and enforce N/A rules)
gt_flat = flatten_and_fix_gt(gt_dict.get(image_id, {}))
# 2) Run prediction; pass gt_flat in (optional) for logging
out, status = process_one_image(img_path, gt_flat, margin_px=margin_px)
if status != "ok":
failures += 1
rows.append({
"image_id": image_id,
"image_path": img_path,
"status": status,
})
print(f"[WARN] {image_id}: {status}")
continue
pred, gt = out # gt should be the flat dict (or subset) returned by process_one_image
# add GT columns + compute accuracy
row = dict(pred)
row["status"] = "ok"
# write GT columns (only non-empty)
for k, v in gt.items():
if v != "" and v is not None:
row[f"gt_{k}"] = v
# score
for k in correct.keys():
if gt.get(k, "") != "": # field exists in GT
total[k] += 1
if str(pred[k]).strip().lower() == str(gt[k]).strip().lower():
correct[k] += 1
rows.append(row)
# Print only incorrect visibility cases
end_vis_wrong = (
gt.get("end_of_line_visible", "") != "" and
pred["end_of_line_visible"] != gt["end_of_line_visible"]
)
start_vis_wrong = (
gt.get("start_of_line_visible", "") != "" and
pred["start_of_line_visible"] != gt["start_of_line_visible"]
)
if end_vis_wrong or start_vis_wrong:
end_loc = pred["end_of_line_location_if_visible"] if pred["end_of_line_visible"] == "yes" else "N/A"
gt_end_loc = gt.get("end_of_line_location_if_visible", "N/A")
start_loc = pred["start_of_line_location_if_visible"] if pred["start_of_line_visible"] == "yes" else "N/A"
gt_start_loc = gt.get("start_of_line_location_if_visible", "N/A")
print(
f"[VIS ERROR] {image_id} | "
f"end: pred={pred['end_of_line_visible']} "
f"(loc={end_loc}) gt={gt.get('end_of_line_visible')} "
f"start: pred={pred['start_of_line_visible']} "
f"(loc={start_loc}) gt={gt.get('start_of_line_visible')} "
)
# Print accuracies
print("\n=== Accuracy (only where GT field exists) ===")
for k in correct.keys():
if total[k] == 0:
print(f"{k}: N/A (no GT)")
else:
acc = correct[k] / total[k]
print(f"{k}: {acc:.4f} ({correct[k]}/{total[k]})")
print(f"\nFailures: {failures}/{len(image_paths)}")
# Save CSV
out_csv = "line_visibility_results.csv"
# gather all possible columns (pred + gt + status)
all_cols = set()
for r in rows:
all_cols.update(r.keys())
all_cols = sorted(all_cols)
with open(out_csv, "w", newline="", encoding="utf-8") as f:
w = csv.DictWriter(f, fieldnames=all_cols)
w.writeheader()
for r in rows:
w.writerow(r)
print(f"\nSaved: {out_csv}")
if __name__ == "__main__":
main()
|