Update : convert coco data format to FLIR, KAIST dataset

MISC/_check_flir.py

@@ -0,0 +1,100 @@
+# Description: Check the FLIR dataset with visualize
+
+from pycocotools.coco import COCO
+import cv2
+import os
+import numpy as np
+
+
+COLOR_PALETTE = np.array([[255, 0, 0], [0, 255, 0], [0, 0, 255], [255, 255, 0], [255, 0, 255], [0, 255, 255], [255, 255, 255]])
+
+def get_category_names(coco):
+    """Get category names from COCO annotation file"""
+    cats = coco.loadCats(coco.getCatIds())
+    return {cat['id']: cat['name'] for cat in cats}
+
+def draw_bbox(image, bbox, category_id, color, category_names):
+    x, y, w, h = [int(b) for b in bbox]
+    cv2.rectangle(image, (x, y), (x+w, y+h), color.tolist(), 2)
+    # Add category label with name
+    label = f"{category_names.get(category_id, f'unknown-{category_id}')}"
+    # Get text size for better positioning
+    (text_width, text_height), baseline = cv2.getTextSize(label, cv2.FONT_HERSHEY_SIMPLEX, 0.5, 2)
+    # Draw background rectangle for text
+    cv2.rectangle(image, (x, y-text_height-baseline-5), (x+text_width, y), color.tolist(), -1)
+    # Draw text
+    cv2.putText(image, label, (x, y-baseline-3), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 0), 2)
+    return image
+
+
+def visualize(coco_path, root, flag):
+    rgb_root = os.path.join(root, flag + '_RGB')
+    thermal_root = os.path.join(root, flag + '_thermal')
+    coco = COCO(coco_path)
+
+    # Get category names from annotation file
+    category_names = get_category_names(coco)
+    print("Categories in dataset:", category_names)
+
+    img_ids = coco.getImgIds()
+    current_idx = 0
+
+    while True:
+        # Get current image info
+        img_info = coco.loadImgs(img_ids[current_idx])[0]
+        img_name = img_info['file_name_RGB']
+        thermal_name = img_info['file_name_IR']
+
+        # Load RGB and thermal images
+        rgb_path = os.path.join(rgb_root, img_name)
+        thermal_path = os.path.join(thermal_root, thermal_name)
+
+        rgb_img = cv2.imread(rgb_path)
+        thermal_img = cv2.imread(thermal_path)
+
+        if rgb_img is None or thermal_img is None:
+            print(f"Error loading images: {img_name}")
+            current_idx = (current_idx + 1) % len(img_ids)
+            continue
+
+        # Get annotations for current image
+        ann_ids = coco.getAnnIds(imgIds=img_info['id'])
+        anns = coco.loadAnns(ann_ids)
+
+        # Draw bounding boxes
+        rgb_vis = rgb_img.copy()
+        thermal_vis = thermal_img.copy()
+
+        for ann in anns:
+            bbox = ann['bbox']
+            category_id = ann['category_id']
+            color = COLOR_PALETTE[category_id % len(COLOR_PALETTE)]
+
+            rgb_vis = draw_bbox(rgb_vis, bbox, category_id, color, category_names)
+            thermal_vis = draw_bbox(thermal_vis, bbox, category_id, color, category_names)
+
+        # Display images side by side
+        combined_img = np.hstack((rgb_vis, thermal_vis))
+        cv2.imshow('FLIR Dataset Visualization (Press q to quit, <- or -> to navigate)', combined_img)
+
+        # Handle keyboard input
+        key = cv2.waitKey(0) & 0xFF
+        if key == ord('q'):  # Quit
+            break
+        elif key == 83 or key == ord('d'):  # Right arrow or 'd'
+            current_idx = (current_idx + 1) % len(img_ids)
+        elif key == 81 or key == ord('a'):  # Left arrow or 'a'
+            current_idx = (current_idx - 1) % len(img_ids)
+
+    cv2.destroyAllWindows()
+
+
+def main():
+    coco_path = '/media/ailab/HDD1/Workspace/dset/Drone-Detection-Benchmark/Source/FLIR-align/annotations/val.json'
+    root = '/media/ailab/HDD1/Workspace/dset/Drone-Detection-Benchmark/Source/FLIR-align/'
+    flag = os.path.basename(coco_path).split('.')[0]
+    visualize(coco_path, root, flag)
+
+
+if __name__ == "__main__":
+    main()

MISC/_check_llvip.py

@@ -0,0 +1,101 @@
+# Description: Check the FLIR dataset with visualize
+
+from pycocotools.coco import COCO
+import cv2
+import os
+import numpy as np
+
+from check_flir import get_category_names, draw_bbox
+COLOR_PALETTE = np.array([[255, 0, 0], [0, 255, 0], [0, 0, 255], [255, 255, 0], [255, 0, 255], [0, 255, 255], [255, 255, 255]])
+
+# def get_category_names(coco):
+#     """Get category names from COCO annotation file"""
+#     cats = coco.loadCats(coco.getCatIds())
+#     return {cat['id']: cat['name'] for cat in cats}
+
+# def draw_bbox(image, bbox, category_id, color, category_names):
+#     x, y, w, h = [int(b) for b in bbox]
+#     cv2.rectangle(image, (x, y), (x+w, y+h), color.tolist(), 2)
+#     # Add category label with name
+#     label = f"{category_names.get(category_id, f'unknown-{category_id}')}"
+#     # Get text size for better positioning
+#     (text_width, text_height), baseline = cv2.getTextSize(label, cv2.FONT_HERSHEY_SIMPLEX, 0.5, 2)
+#     # Draw background rectangle for text
+#     cv2.rectangle(image, (x, y-text_height-baseline-5), (x+text_width, y), color.tolist(), -1)
+#     # Draw text
+#     cv2.putText(image, label, (x, y-baseline-3), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 0), 2)
+#     return image
+
+
+def visualize(coco_path, root, flag):
+    if flag == 'val':
+        flag = 'test'
+    rgb_root = os.path.join(root, 'visible')
+    thermal_root = os.path.join(root,'infrared')
+    coco = COCO(coco_path)
+
+    # Get category names from annotation file
+    category_names = get_category_names(coco)
+    print("Categories in dataset:", category_names)
+
+    img_ids = coco.getImgIds()
+    current_idx = 0
+
+    while True:
+        # Get current image info
+        img_info = coco.loadImgs(img_ids[current_idx])[0]
+        img_name = img_info['file_name']
+        thermal_name = img_info['file_name']
+
+        # Load RGB and thermal images
+        rgb_path = os.path.join(rgb_root,flag,  img_name)
+        thermal_path = os.path.join(thermal_root,flag, thermal_name)
+
+        rgb_img = cv2.imread(rgb_path)
+        thermal_img = cv2.imread(thermal_path)
+
+        if rgb_img is None or thermal_img is None:
+            print(f"Error loading images: {img_name}")
+            current_idx = (current_idx + 1) % len(img_ids)
+            continue
+
+        # Get annotations for current image
+        ann_ids = coco.getAnnIds(imgIds=img_info['id'])
+        anns = coco.loadAnns(ann_ids)
+
+        # Draw bounding boxes
+        rgb_vis = rgb_img.copy()
+        thermal_vis = thermal_img.copy()
+
+        for ann in anns:
+            bbox = ann['bbox']
+            category_id = ann['category_id']
+            color = COLOR_PALETTE[category_id % len(COLOR_PALETTE)]
+            rgb_vis = draw_bbox(rgb_vis, bbox, category_id, color, category_names)
+            thermal_vis = draw_bbox(thermal_vis, bbox, category_id, color, category_names)
+
+        # Display images side by side
+        combined_img = np.hstack((rgb_vis, thermal_vis))
+        cv2.imshow('FLIR Dataset Visualization (Press q to quit, <- or -> to navigate)', combined_img)
+
+        # Handle keyboard input
+        key = cv2.waitKey(0) & 0xFF
+        if key == ord('q'):  # Quit
+            break
+        elif key == 83 or key == ord('d'):  # Right arrow or 'd'
+            current_idx = (current_idx + 1) % len(img_ids)
+        elif key == 81 or key == ord('a'):  # Left arrow or 'a'
+            current_idx = (current_idx - 1) % len(img_ids)
+
+    cv2.destroyAllWindows()
+
+
+def main():
+    coco_path = '/media/ailab/HDD1/Workspace/dset/Drone-Detection-Benchmark/Source/LLVIP/coco_annotations/train.json'
+    root = '/media/ailab/HDD1/Workspace/dset/Drone-Detection-Benchmark/Source/LLVIP/'
+    flag = os.path.basename(coco_path).split('.')[0]
+    visualize(coco_path, root, flag)
+
+
+if __name__ == "__main__":
+    main()

MISC/_process_flir.py

@@ -0,0 +1,54 @@
+
+from pycocotools.coco import COCO
+# val_coco = COCO('/media/ailab/HDD1/Workspace/dset/Drone-Detection-Benchmark/Source/Annotations/FLIR_coco/val.json')
+# num_images = len(coco_annotation.getImgIds())
+# print("Number of images:", num_images)
+# # %%
+# train_coco = COCO('/media/ailab/HDD1/Workspace/dset/Drone-Detection-Benchmark/Source/Annotations/FLIR_coco/train.json')
+# num_imgs = len(train_coco.getImgIds())
+# print("Number of images:", num_imgs)
+# # %%
+# import glob
+
+# search_root = '/media/ailab/HDD1/Workspace/dset/Drone-Detection-Benchmark/Source/FLIR-align'
+# id = train_coco.getImgIds()[0]
+# img = train_coco.loadImgs(id)[0]
+
+# # %%
+# rgb_name = img['file_name_RGB']
+# thermal_name = img['file_name_IR']
+# rgb_files = glob.glob(search_root + '/*/' + rgb_name)
+# thermal_files = glob.glob(search_root + '/*/' + thermal_name)
+
+# %%
+
+# %%
+import shutil
+import glob
+def copy_files(coco_path, isTrain):
+    search_root = '/media/ailab/HDD1/Workspace/dset/Drone-Detection-Benchmark/Source/FLIR-align'
+    coco = COCO(coco_path)
+    if isTrain:
+        rgb_root = '/media/ailab/HDD1/Workspace/dset/Drone-Detection-Benchmark/Source/FLIR-align/train_RGB'
+        thermal_root = '/media/ailab/HDD1/Workspace/dset/Drone-Detection-Benchmark/Source/FLIR-align/train_thermal'
+    else:
+        rgb_root = '/media/ailab/HDD1/Workspace/dset/Drone-Detection-Benchmark/Source/FLIR-align/val_RGB'
+        thermal_root = '/media/ailab/HDD1/Workspace/dset/Drone-Detection-Benchmark/Source/FLIR-align/val_thermal'
+
+    for id in coco.getImgIds():
+        img = coco.loadImgs(id)[0]
+        rgb_name = img['file_name_RGB']
+        thermal_name = img['file_name_IR']
+        rgb_file = glob.glob(search_root + '/JPEGImages/' + rgb_name)[0]
+        thermal_file = glob.glob(search_root + '/JPEGImages/' + thermal_name)[0]
+        rgb_filename = rgb_file.split('/')[-1]
+        thermal_filename = thermal_file.split('/')[-1]
+
+        shutil.copy2(rgb_file, rgb_root + '/' + rgb_filename)
+        shutil.copy2(thermal_file, thermal_root + '/' + thermal_filename)
+        print(f"Copy {rgb_file} to {rgb_root + '/' + rgb_filename}  |   {thermal_file} to {thermal_root + '/' + thermal_filename}")
+
+
+
+# copy_files('/media/ailab/HDD1/Workspace/dset/Drone-Detection-Benchmark/Source/Annotations/FLIR_coco/val.json', False)
+copy_files('/media/ailab/HDD1/Workspace/dset/Drone-Detection-Benchmark/Source/Annotations/FLIR_coco/train.json', True)