Datasets:
File size: 4,759 Bytes
ada18d0 | 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 | import json
import os
import matplotlib.pyplot as plt
from matplotlib.patches import Patch
from matplotlib.patches import Polygon
import numpy as np
from PIL import Image
from pycocotools.coco import COCO
from pycocotools.cocoeval import COCOeval
xp_images = ["Exp6_Cam1_Stage01.jpg", "Exp6_Cam1_Stage03.jpg", "Exp6_Cam1_Stage05.jpg",
"Exp17_Cam1_Stage01.jpg", "Exp17_Cam1_Stage04.jpg", "Exp17_Cam1_Stage05.jpg",
"Exp30_Cam1_Stage01.jpg", "Exp30_Cam1_Stage05.jpg", "Exp30_Cam1_Stage07.jpg",
"Exp21_Cam2_Stage01.jpg", "Exp21_Cam2_Stage03.jpg", "Exp21_Cam2_Stage05.jpg",
"Exp18_Cam3_Stage01.jpg", "Exp18_Cam3_Stage04.jpg", "Exp18_Cam3_Stage06.jpg"]
coco_gt_path = "instance_segmentation/ground_truth_15imgs_coco.json"
multiannotation_path = "instance_segmentation/multiannotator_15imgs.json"
img_folder = "images"
binary_folder = "binary_segmentation"
if __name__ == "__main__":
coco = COCO(coco_gt_path)
# Sanity check: display the ground truth of one image
img_info = coco.imgs[1]
anns = coco.loadAnns(coco.getAnnIds(imgIds=img_info["id"]))
print(f"{img_info['file_name']}: {len(anns)} annotations")
plt.figure(figsize=(12, 8))
plt.imshow(Image.open(os.path.join(img_folder, img_info["file_name"])))
coco.showAnns(anns, draw_bbox=True)
plt.title(img_info["file_name"])
plt.axis("off")
plt.tight_layout()
plt.savefig("example_gt.png", dpi=100)
plt.show()
# Same image, annotated by one of the annotators
with open(multiannotation_path) as f:
multiannotations = json.load(f)
annotation = multiannotations[img_info["file_name"]][1] # index 1 is the second annotator, index 0 is the first annotator, etc
print(f"{img_info['file_name']}: annotator {annotation['user_id']}, "
f"{len(annotation['polygons'])} polygons")
fig, ax = plt.subplots(figsize=(12, 8))
ax.imshow(Image.open(os.path.join(img_folder, img_info["file_name"])))
for poly in annotation["polygons"]:
pts = [(p["x"], p["y"]) for p in poly["polygon_pts"]]
ax.add_patch(Polygon(pts, closed=True, fill=True, alpha=0.4,
facecolor="tab:orange", edgecolor="tab:orange", linewidth=2))
ax.set_title(f"{img_info['file_name']} - annotator {annotation['user_id']}")
ax.axis("off")
fig.tight_layout()
fig.savefig("example_annotator.png", dpi=100)
plt.show()
# Evaluate the annotator against the ground truth with COCO metrics (this image only).
# User-reported confidence is used as the bbox/segmentation score.
detections = []
for poly in annotation["polygons"]:
flat = [c for p in poly["polygon_pts"] for c in (p["x"], p["y"])]
xs, ys = flat[0::2], flat[1::2]
detections.append({
"image_id": img_info["id"], "category_id": 1, "segmentation": [flat],
"bbox": [min(xs), min(ys), max(xs) - min(xs), max(ys) - min(ys)],
"score": poly.get("polygon_confidence") or 1.0,
})
coco_dt = coco.loadRes(detections)
for iou_type in ("segm", "bbox"):
print(f"\n=== {iou_type} ===")
coco_eval = COCOeval(coco, coco_dt, iou_type)
coco_eval.params.imgIds = [img_info["id"]]
coco_eval.evaluate()
coco_eval.accumulate()
coco_eval.summarize()
# Binary (foreground/background) IoU between the annotator and ground truth masks
name = img_info["file_name"]
gt_mask = np.array(Image.open(os.path.join(binary_folder, f"{name}_GT.png"))).astype(bool)
annotator_mask = np.array(
Image.open(os.path.join(binary_folder, f"{name}_{annotation['user_id']}.png"))).astype(bool)
intersection = np.logical_and(gt_mask, annotator_mask).sum()
union = np.logical_or(gt_mask, annotator_mask).sum()
print(f"\nBinary IoU ({name}, annotator {annotation['user_id']}): {intersection / union:.3f}")
# Overlay both masks on the image: agreement in green, GT only in red, annotator only in blue
overlay = np.zeros((*gt_mask.shape, 4))
overlay[gt_mask & annotator_mask] = (0, 0.8, 0, 0.5)
overlay[gt_mask & ~annotator_mask] = (1, 0, 0, 0.5)
overlay[~gt_mask & annotator_mask] = (0, 0.3, 1, 0.5)
fig, ax = plt.subplots(figsize=(12, 8))
ax.imshow(Image.open(os.path.join(img_folder, name)))
ax.imshow(overlay)
ax.legend(handles=[Patch(color=(0, 0.8, 0), label="both"),
Patch(color=(1, 0, 0), label="GT only"),
Patch(color=(0, 0.3, 1), label="annotator only")], loc="lower right")
ax.set_title(f"{name} - GT vs annotator {annotation['user_id']} (IoU {intersection / union:.3f})")
ax.axis("off")
fig.tight_layout()
fig.savefig("example_binary_masks.png", dpi=100)
plt.show()
|