Dataset loading script (#5)

- Provide bash script for simple dataset access (547112a51577caf8456e69a10c9c5154020d9ace)
- Use curl instead of wget for better platform compatibility (ef2fb675c2491908c8b0fd1554e4a5bf7c96ea9b)
- Convert download script to Python (3f19446e290ae71aa3a74532567c0505ae937946)
- Remove bash script (1f3d066c0cf9912b1784812f07d3fc201ea60794)
- Reduce memory usage and improve storage efficiency during dataset processing (3ed547e2b80f2a3a6f1c6f79d1739f6309f010b8)


Co-authored-by: Matthew Thompson <thompsonmj@users.noreply.huggingface.co>

download.py

@@ -0,0 +1,154 @@
+import os
+import requests
+import time
+import zipfile
+import glob
+from hashlib import md5
+import concurrent.futures
+
+base_url = "https://huggingface.co/datasets/imageomics/KABR/resolve/main/KABR"
+
+"""
+To extend the dataset, add additional animals and parts ranges to the list and dictionary below.
+"""
+
+animals = ["giraffes", "zebras_grevys", "zebras_plains"]
+
+animal_parts_range = {
+    "giraffes": ("aa", "ad"),
+    "zebras_grevys": ("aa", "am"),
+    "zebras_plains": ("aa", "al"),
+}
+
+dataset_prefix = "dataset/image/"
+
+# Define the static files that are not dependent on the animals list
+static_files = [
+    "README.txt",
+    "annotation/classes.json",
+    "annotation/distribution.xlsx",
+    "annotation/train.csv",
+    "annotation/val.csv",
+    "configs/I3D.yaml",
+    "configs/SLOWFAST.yaml",
+    "configs/X3D.yaml",
+    "dataset/image2video.py",
+    "dataset/image2visual.py",
+]
+
+def generate_part_files(animal, start, end):
+    start_a, start_b = ord(start[0]), ord(start[1])
+    end_a, end_b = ord(end[0]), ord(end[1])
+    return [
+        f"{dataset_prefix}{animal}_part_{chr(a)}{chr(b)}"
+        for a in range(start_a, end_a + 1)
+        for b in range(start_b, end_b + 1)
+    ]
+
+# Generate the part files for each animal
+part_files = [
+    part
+    for animal, (start, end) in animal_parts_range.items()
+    for part in generate_part_files(animal, start, end)
+]
+
+archive_md5_files = [f"{dataset_prefix}{animal}_md5.txt" for animal in animals]
+
+files = static_files + archive_md5_files + part_files
+
+def progress_bar(iteration, total, message, bar_length=50):
+    progress = (iteration / total)
+    bar = '=' * int(round(progress * bar_length) - 1)
+    spaces = ' ' * (bar_length - len(bar))
+    message = f'{message:<100}'
+    print(f'[{bar + spaces}] {int(progress * 100)}% {message}', end='\r', flush=True)
+
+    if iteration == total:
+        print()
+
+# Directory to save files
+save_dir = "KABR_files"
+
+# Loop through each relative file path
+
+print(f"Downloading the Kenyan Animal Behavior Recognition (KABR) dataset ...")
+
+total = len(files)
+for i, file_path in enumerate(files):
+    # Construct the full URL
+    save_path = os.path.join(save_dir, file_path)
+    
+    if os.path.exists(save_path):
+        print(f"File {save_path} already exists. Skipping download.")
+        continue 
+    
+    full_url = f"{base_url}/{file_path}"
+    
+    # Create the necessary directories based on the file path
+    os.makedirs(os.path.join(save_dir, os.path.dirname(file_path)), exist_ok=True)
+    
+    # Download the file and save it with the preserved file path
+    response = requests.get(full_url)
+    with open(save_path, 'wb') as file:
+        file.write(response.content)
+    
+    progress_bar(i+1, total, f"downloaded: {save_path}")
+    
+print("Download of repository contents completed.")
+
+print(f"Concatenating split files into a full archive for {animals} ...")
+
+def concatenate_files(animal):
+    print(f"Concatenating files for {animal} ...")
+    part_files_pattern = f"{save_dir}/dataset/image/{animal}_part_*"
+    part_files = sorted(glob.glob(part_files_pattern))
+    if part_files:
+        with open(f"{save_dir}/dataset/image/{animal}.zip", 'wb') as f_out:
+            for f_name in part_files:
+                with open(f_name, 'rb') as f_in:
+                    # Read and write in chunks
+                    CHUNK_SIZE = 8*1024*1024 # 8MB
+                    for chunk in iter(lambda: f_in.read(CHUNK_SIZE), b""):
+                        f_out.write(chunk) 
+                # Delete part files as they are concatenated              
+                os.remove(f_name)
+        print(f"Archive for {animal} concatenated.")
+    else:
+        print(f"No part files found for {animal}.")
+
+with concurrent.futures.ThreadPoolExecutor() as executor:
+    executor.map(concatenate_files, animals)
+
+def compute_md5(file_path):
+    hasher = md5()
+    with open(file_path, 'rb') as f:
+        CHUNK_SIZE = 8*1024*1024 # 8MB
+        for chunk in iter(lambda: f.read(CHUNK_SIZE), b""):
+            hasher.update(chunk)
+    return hasher.hexdigest()
+    
+def verify_and_extract(animal):
+    print(f"Confirming data integrity for {animal}.zip ...")
+    zip_md5 = compute_md5(f"{save_dir}/dataset/image/{animal}.zip")
+    
+    with open(f"{save_dir}/dataset/image/{animal}_md5.txt", 'r') as file:
+        expected_md5 = file.read().strip().split()[0]
+    
+    if zip_md5 == expected_md5:
+        print(f"MD5 sum for {animal}.zip is correct.")
+
+        print(f"Extracting {animal}.zip ...")
+        with zipfile.ZipFile(f"{save_dir}/dataset/image/{animal}.zip", 'r') as zip_ref:
+            zip_ref.extractall(f"{save_dir}/dataset/image/")
+        print(f"{animal}.zip extracted.")
+        print(f"Cleaning up for {animal} ...")
+        os.remove(f"{save_dir}/dataset/image/{animal}.zip")
+        os.remove(f"{save_dir}/dataset/image/{animal}_md5.txt")
+    else:
+        print(f"MD5 sum for {animal}.zip is incorrect. Expected: {expected_md5}, but got: {zip_md5}.")
+        print("There may be data corruption. Please try to download and reconstruct the data again or reach out to the corresponding authors for assistance.")
+
+with concurrent.futures.ThreadPoolExecutor() as executor:
+    executor.map(verify_and_extract, animals)
+
+print("Download script finished.")