utils/upload-to-dataset.py

from datasets import Dataset, Features, Value, Image
from huggingface_hub import HfApi
import os
from collections import defaultdict
import pandas as pd

# Define the path to your images
IMAGE_DIR = "../../background-removal-arena-v0/train/data/resized"

# Define the dataset features with dedicated columns for each model
features = Features({
    "original_image": Image(),  # Original image feature
    "clipdrop_image": Image(),  # Clipdrop segmented image
    "bria_image": Image(),      # Bria segmented image
    "photoroom_image": Image(), # Photoroom segmented image
    "removebg_image": Image(),  # RemoveBG segmented image
    "original_filename": Value("string")  # Original filename
})

# Load image paths and metadata
data = defaultdict(lambda: {
    "clipdrop_image": None,
    "bria_image": None,
    "photoroom_image": None,
    "removebg_image": None
})

# Walk into the web-original-images folder
web_original_images_dir = os.path.join(IMAGE_DIR, "web-original-images")
for root, _, files in os.walk(web_original_images_dir):
    for f in files:
        if f.endswith(('.png', '.jpg', '.jpeg')):
            original_image_path = os.path.join(root, f)
            data[f]["original_image"] = original_image_path
            data[f]["original_filename"] = f

            # Check for corresponding images in other directories
            for source in ["clipdrop", "bria", "photoroom", "removebg"]:
                # Check for processed images ending in .png or .jpg
                for ext in ['.png', '.jpg']:
                    processed_image_filename = os.path.splitext(f)[0] + ext
                    source_image_path = os.path.join(IMAGE_DIR, source, processed_image_filename)
    
                    if os.path.exists(source_image_path):
                        data[f][f"{source}_image"] = source_image_path
                        break  # Stop checking other extensions if a file is found

# Convert the data to a dictionary of lists
dataset_dict = {
    "original_image": [],
    "clipdrop_image": [],
    "bria_image": [],
    "photoroom_image": [],
    "removebg_image": [],
    "original_filename": []
}

for filename, entry in data.items():
    if "original_image" in entry:
        dataset_dict["original_image"].append(entry["original_image"])
        dataset_dict["clipdrop_image"].append(entry["clipdrop_image"])
        dataset_dict["bria_image"].append(entry["bria_image"])
        dataset_dict["photoroom_image"].append(entry["photoroom_image"])
        dataset_dict["removebg_image"].append(entry["removebg_image"])
        dataset_dict["original_filename"].append(filename)

# Save the data dictionary to a CSV file for inspection
df = pd.DataFrame.from_dict(dataset_dict)
df.to_csv("image_data.csv", index=False)

# Create a Dataset
dataset = Dataset.from_dict(dataset_dict, features=features)

# Push the dataset to Hugging Face Hub
api = HfApi()
dataset.push_to_hub("bgsys/background-removal-arena-test", token=api.token)