Object Detection
ultralytics
English
biology
CV
images
animals
YOLO
fine-tuned
zebra
giraffe
onager
dog
Instructions to use imageomics/mmla with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- ultralytics
How to use imageomics/mmla with ultralytics:
# Couldn't find a valid YOLO version tag. # Replace XX with the correct version. from ultralytics import YOLOvXX model = YOLOvXX.from_pretrained("imageomics/mmla") source = 'http://images.cocodataset.org/val2017/000000039769.jpg' model.predict(source=source, save=True) - Notebooks
- Google Colab
- Kaggle
| import os | |
| import random | |
| from pathlib import Path | |
| import shutil | |
| ####################################################### | |
| # CONFIGURATION SECTION - MODIFY THESE VALUES | |
| ####################################################### | |
| # Define source directories for each location | |
| SOURCE_DIRS = { | |
| 'location_1': 'mpala', # REPLACE WITH YOUR ACTUAL PATH | |
| 'location_2': 'opc', # REPLACE WITH YOUR ACTUAL PATH | |
| 'location_3': 'wilds' # REPLACE WITH YOUR ACTUAL PATH | |
| } | |
| # Destination directory | |
| DEST_DIR = "/data" # REPLACE WITH YOUR ACTUAL PATH | |
| # Define your class labels | |
| CLASS_LABELS = { | |
| 0: "Zebra", | |
| 1: "Giraffe", | |
| 2: "Onager", | |
| 3: "Dog", | |
| } | |
| # Sampling rate (adjust as needed - higher values mean fewer frames) | |
| SAMPLING_RATE = 10 | |
| # Define the splits (train/test) for the 70/30 strategy | |
| splits = { | |
| 'train': { | |
| 'location_3': { | |
| 'session_1': ['DJI_0034', 'DJI_0035_part1'], # African Painted Dog (70%) | |
| 'session_2': ['P0140018'], # Giraffe (70%) | |
| 'session_3': ['P0100010', 'P0110011', 'P0080008', 'P0090009'], # Persian Onanger (70%) | |
| }, | |
| 'location_1': { | |
| 'session_1': ['DJI_0001', 'DJI_0002'], # Giraffe | |
| 'session_2': ['DJI_0005', 'DJI_0006'], # Plains zebra | |
| 'session_3': ['DJI_0068', 'DJI_0069'], # Grevy's zebra | |
| 'session_4': ['DJI_0142', 'DJI_0143', 'DJI_0144'], # Grevy's zebra | |
| 'session_5': ['DJI_0206', 'DJI_0208'], # Mixed species | |
| }, | |
| 'location_2': { | |
| 'session_1': ['P0800081', 'P0830086', 'P0840087', 'P0870091'], # Plains zebra | |
| 'session_2': ['P0910095'], # Plains zebra | |
| } | |
| }, | |
| 'test': { | |
| 'location_3': { | |
| 'session_1': ['DJI_0035_part2'], # African Painted Dog (30%) | |
| 'session_3': ['P0070007', 'P0160016', 'P0120012'], # Persian Onanger (30%) | |
| 'session_2': ['P0150019'], # Giraffe (30%) | |
| 'session_4': ['P0070010'], # Grevy's Zebra (100%) | |
| }, | |
| 'location_1': { | |
| 'session_3': ['DJI_0070', 'DJI_0071'], # Grevy's zebra | |
| 'session_4': ['DJI_0145', 'DJI_0146', 'DJI_0147'], # Grevy's zebra | |
| 'session_5': ['DJI_0210', 'DJI_0211'], # Mixed species | |
| }, | |
| 'location_2': { | |
| 'session_1': ['P0860090'], # Plains zebra | |
| 'session_2': ['P0940098'], # Plains zebra | |
| } | |
| } | |
| } | |
| ####################################################### | |
| # SCRIPT CODE - DO NOT MODIFY UNLESS NECESSARY | |
| ####################################################### | |
| # Create destination directories | |
| for split in ['train', 'test']: | |
| os.makedirs(f"{DEST_DIR}/images/{split}", exist_ok=True) | |
| os.makedirs(f"{DEST_DIR}/labels/{split}", exist_ok=True) | |
| def find_images_in_directory(dir_path): | |
| """Find all image files in a directory""" | |
| try: | |
| return [f for f in os.listdir(dir_path) | |
| if f.endswith(('.jpg', '.png', '.jpeg')) and os.path.isfile(dir_path / f)] | |
| except (FileNotFoundError, NotADirectoryError, PermissionError) as e: | |
| print(f"Error accessing {dir_path}: {e}") | |
| return [] | |
| def find_partitions(session_path): | |
| """Find partition directories in a session""" | |
| try: | |
| return [d for d in os.listdir(session_path) | |
| if os.path.isdir(session_path / d) and d.startswith('partition_')] | |
| except (FileNotFoundError, NotADirectoryError, PermissionError) as e: | |
| print(f"Error accessing {session_path}: {e}") | |
| return [] | |
| def find_video_images(session_path, video_name): | |
| """ | |
| Find all images for a specific video in all partitions or video directory | |
| Returns a list of tuples: (image_path, image_name, partition_name) | |
| """ | |
| all_images = [] | |
| # First, check if the video is directly a directory | |
| video_path = session_path / video_name | |
| if os.path.isdir(video_path): | |
| # Check for partitions within video directory | |
| partitions = find_partitions(video_path) | |
| if partitions: | |
| # If partitions exist in video directory | |
| for partition in partitions: | |
| partition_path = video_path / partition | |
| images = find_images_in_directory(partition_path) | |
| all_images.extend([(partition_path, img, partition) for img in images]) | |
| else: | |
| # Check for direct images in video directory (no partitions) | |
| images = find_images_in_directory(video_path) | |
| all_images.extend([(video_path, img, "") for img in images]) | |
| # Also check for partitions directly in session directory | |
| partitions = find_partitions(session_path) | |
| for partition in partitions: | |
| partition_path = session_path / partition | |
| # Look for images matching this video name pattern | |
| for img in find_images_in_directory(partition_path): | |
| # Check if image filename contains this video name | |
| if video_name in img: | |
| all_images.append((partition_path, img, partition)) | |
| return all_images | |
| # Process each location and session | |
| for split_name, locations in splits.items(): | |
| for location_name, sessions in locations.items(): | |
| # Get the source directory for this location | |
| if location_name not in SOURCE_DIRS: | |
| print(f"Warning: No source directory defined for {location_name}. Skipping.") | |
| continue | |
| location_source_dir = Path(SOURCE_DIRS[location_name]) | |
| for session_name, video_info in sessions.items(): | |
| session_path = location_source_dir / session_name | |
| if not os.path.exists(session_path): | |
| print(f"Warning: Session path {session_path} does not exist. Skipping.") | |
| continue | |
| # Get all videos in this session | |
| if isinstance(video_info, bool) and video_info: | |
| # Use all videos in the session - detect them from directories or video files | |
| try: | |
| # First check for video directories | |
| videos = [v for v in os.listdir(session_path) | |
| if os.path.isdir(session_path / v) and not v.startswith('partition_')] | |
| # If no video directories, try to infer from partition files | |
| if not videos: | |
| partitions = find_partitions(session_path) | |
| if partitions: | |
| # Get all images in first partition to extract video names | |
| first_partition = session_path / partitions[0] | |
| all_imgs = find_images_in_directory(first_partition) | |
| # Extract potential video names from image filenames | |
| videos = list(set([img.split('_')[0] for img in all_imgs if '_' in img])) | |
| except (FileNotFoundError, NotADirectoryError) as e: | |
| print(f"Warning: Could not list directory {session_path}: {e}") | |
| continue | |
| else: | |
| # Use specific videos | |
| videos = video_info | |
| # Process each video | |
| for video in videos: | |
| print(f"Processing {location_name}/{session_name}/{video}...") | |
| # Find all images for this video (in all partitions) | |
| frame_info = find_video_images(session_path, video) | |
| if not frame_info: | |
| print(f"Warning: No frames found for {video} in {session_name}") | |
| continue | |
| # Sort frames by name to ensure temporal order | |
| frame_info.sort(key=lambda x: x[1]) | |
| # Sample frames at regular intervals | |
| sampled_frame_info = frame_info[::SAMPLING_RATE] | |
| # Copy sampled frames and labels to destination | |
| for frame_dir, frame_name, partition in sampled_frame_info: | |
| # Create a path component for the partition if it exists | |
| partition_str = "" if partition == "" else f"_{partition}" | |
| # Copy image | |
| src_img = frame_dir / frame_name | |
| dest_img_name = f"{location_name}_{session_name}_{video}{partition_str}_{frame_name}" | |
| dest_img = Path(DEST_DIR) / "images" / split_name / dest_img_name | |
| try: | |
| shutil.copy(src_img, dest_img) | |
| except (FileNotFoundError, IOError) as e: | |
| print(f"Error copying image {src_img}: {e}") | |
| continue | |
| # Handle different possible label locations | |
| label_name = frame_name.replace('.jpg', '.txt').replace('.png', '.txt').replace('.jpeg', '.txt') | |
| # Possible label locations (in order of priority) | |
| possible_label_paths = [ | |
| # 1. Same directory as image | |
| frame_dir / label_name, | |
| # 2. Labels subdirectory in partition | |
| frame_dir / "labels" / label_name, | |
| # 3. Labels directory parallel to partition with same structure | |
| session_path / "labels" / partition / label_name, | |
| # 4. Flat labels directory for session | |
| session_path / "labels" / label_name, | |
| # 5. In video directory (if it exists) | |
| session_path / video / "labels" / label_name, | |
| ] | |
| src_label = None | |
| for label_path in possible_label_paths: | |
| if os.path.exists(label_path): | |
| src_label = label_path | |
| break | |
| if src_label: | |
| dest_label_name = dest_img_name.replace('.jpg', '.txt').replace('.png', '.txt').replace('.jpeg', '.txt') | |
| dest_label = Path(DEST_DIR) / "labels" / split_name / dest_label_name | |
| try: | |
| shutil.copy(src_label, dest_label) | |
| except (FileNotFoundError, IOError) as e: | |
| print(f"Error copying label {src_label}: {e}") | |
| else: | |
| print(f"Warning: No label found for {src_img}") | |
| print("Dataset split completed successfully!") | |
| # Create dataset.yaml file | |
| def create_dataset_yaml(): | |
| with open(f"{DEST_DIR}/dataset.yaml", "w") as f: | |
| f.write(f"# YOLOv11 dataset config\n") | |
| f.write(f"path: {os.path.abspath(DEST_DIR)} # dataset root dir\n") | |
| f.write(f"train: images/train # train images\n") | |
| f.write(f"val: images/train # validation uses train images\n") | |
| f.write(f"test: images/test # test images\n\n") | |
| f.write(f"# Classes\n") | |
| f.write(f"names:\n") | |
| for class_id, class_name in CLASS_LABELS.items(): | |
| f.write(f" {class_id}: {class_name}\n") | |
| create_dataset_yaml() | |
| # Analyze the distribution | |
| stats = {"train": {}, "test": {}} | |
| for split in ['train', 'test']: | |
| # Count images by location | |
| locations = {} | |
| species_count = {} | |
| # Get all images in this split | |
| img_dir = Path(DEST_DIR) / "images" / split | |
| if not os.path.exists(img_dir): | |
| print(f"Warning: Directory {img_dir} does not exist.") | |
| continue | |
| total_count = 0 | |
| for img in os.listdir(img_dir): | |
| parts = img.split('_') | |
| if len(parts) < 2: | |
| continue | |
| location = parts[0] | |
| session = parts[1] | |
| # Count by location | |
| if location not in locations: | |
| locations[location] = 0 | |
| locations[location] += 1 | |
| # Extract species information if possible | |
| species_key = f"{location}_{session}" | |
| if species_key not in species_count: | |
| species_count[species_key] = 0 | |
| species_count[species_key] += 1 | |
| # Increment total | |
| total_count += 1 | |
| stats[split]["total"] = total_count | |
| stats[split]["locations"] = locations | |
| stats[split]["species"] = species_count | |
| # Print stats | |
| for split, data in stats.items(): | |
| print(f"\n{split.upper()} set:") | |
| print(f"Total images: {data['total']}") | |
| print("Distribution by location:") | |
| for loc, count in data["locations"].items(): | |
| percentage = (count/data['total']*100) if data['total'] > 0 else 0 | |
| print(f" - {loc}: {count} ({percentage:.1f}%)") | |
| print("\nDistribution by location_session:") | |
| for species_key, count in data["species"].items(): | |
| percentage = (count/data['total']*100) if data['total'] > 0 else 0 | |
| print(f" - {species_key}: {count} ({percentage:.1f}%)") | |
| print("\nOverall train/test ratio:", | |
| f"{stats['train']['total'] / (stats['train']['total'] + stats['test']['total']):.1%}", | |
| f"/ {stats['test']['total'] / (stats['train']['total'] + stats['test']['total']):.1%}") |