Ai Background Removal | AI Final Year Project Report

Complete Technical Paper Content — Ai Background Removal

1. Executive Abstract, Problem Formulation & Background

Jev is a proprietary artificial intelligence model developed by TypeSafe AI, a San Francisco–based company founded in 2024. It was released in limited early access on 15 September 2026, alongside the announcement of a US$40 million seed round led by DCVC. Jev is a discriminative model: it is used for classification or regression, not generation. It does not generate natural-language text. It instead returns typed values together with probability estimates and confidence scores. Its output is intended to be consumed directly by software rather than read by a person. Unlike classical discriminative models, Jev does not require data-specific training. Like generative pretrained transformer models, it can perform zero-shot learning of input for categories provided at inference time, based on contextual information provided. TypeSafe describes Jev as the first of a class of models it calls "System One models", a name the company says comes from fast, intuitive System 1 thinking as popularized by Daniel Kahneman.

A state is an entry. An entry is a string, JSON object, or array of string. Examples:A question is a JSON object. Jev allows three types of questions, all multiple-choice.

Schematically: If the state is a JSON object, the question can also refer to attributes of the state. For example:

On November 17, 2023, OpenAI's board of directors ousted co-founder and chief executive Sam Altman. In an official post on the company's website, it was stated that "the board no longer has confidence in his ability to continue leading OpenAI". The removal was predicated by employee concerns, led by co-founder Ilya Sutskever and chief technology officer Mira Murati, about Altman's handling of artificial intelligence safety, commercialization, and allegations of lying and abusive behavior. Altman was reinstated on November 22 after pressure from employees and investors, with negotiations mediated by Microsoft CEO Satya Nadella. The day of the removal, Greg Brockman and Jakub Pachocki resigned in support of Altman. The board of directors appointed Murati as interim CEO, and appointed former Twitch CEO Emmett Shear as CEO on November 19. On November 20, a letter signed by 745 of OpenAI's 770 employees threatened mass resignations if the board does not resign; Sutskever was a signatory, defecting from the board and publicly apologizing for his participation in the board's previous actions. The same day, Microsoft appointed Altman, Brockman, Pachocki, and other ex-OpenAI scientists to a new AI research team at Microsoft. The removal and subsequent reinstatement caused widespread reactions, including impacts felt in the financial markets and technology sector. Microsoft, a partner of OpenAI, received little notice of the removal and experienced a drop in the share price of its stock. The removal also promoted interest in investigations from regulatory agencies.

OpenAI is an artificial intelligence firm founded in December 2015 as a non-profit entity. The for-profit division of the organization released ChatGPT in November 2022, contributing to a resurgence in generative artificial intelligence funding. The board of directors of the controlling non-profit formerly comprised chief scientist Ilya Sutskever, as well as Adam D'Angelo, chief executive of Quora, entrepreneur Tasha McCauley, and Helen Toner, strategy director for the Center for Security and Emerging Technology. As of October 2023, the company is valued at US$80 billion and was set to bring in US$1 billion in revenue. Altman has described OpenAI's relationship with Microsoft as the "best bromance in tech". OpenAI is uniquely structured, an intentional decision to avoid investor control. A board of directors controls the non-profit OpenAI, Inc. The non-profit owns and controls a for-profit company itself controlling a capped-profit company, OpenAI Global, LLC and a holding company owned by employees and other investors. The holding company is the majority owner of OpenAI Global, LLC.; Microsoft owns a minority stake in the capped-profit company. OpenAI's bylaws, enacted in January 2016, allow a majority of its board of directors to remove any director without prior warning or a formal meeting with written consent.

Sam Altman is a co-founder of OpenAI and its former chief executive; Altman took over the company following co-chair Elon Musk's resignation in 2018. Under Altman, OpenAI has shifted to becoming a for-profit entity. Altman is credited with convincing Microsoft chief executive Satya Nadella with investing US$10 billion in cash and computing credits into OpenAI and leading several tender offer transactions that tripled the company's valuation. Altman testified before the United States Congress speaking critically of artificial intelligence and appeared at the 2023 AI Safety Summit. In the days leading up to his removal, Altman made several public appearances, announcing the GPT-4 Turbo platform at OpenAI's DevDay conference, attending APEC United States 2023, and speaking at an event related to Burning Man.

On November 17, 2023, at approximately noon PST, OpenAI's board of directors ousted Altman effective immediately following a "deliberative review process". The board concluded that Altman was not "consistently candid in his communications". Altman was informed of his removal five to ten minutes before it occurred on a Google Meet while watching the Las Vegas Grand Prix. Within thirty minutes, Sutskever invited OpenAI chairman and president Greg Brockman to a Google Meet to inform him of Altman's removal. According to an internal memo obtained by Axios, the removal was not due to "malfeasance", and OpenAI chief executive Emmett Shear denied accusations that the removal was due to disagreements. The board publicly announced Altman's removal thirty minutes later. Chief Technology Officer Mira Murati was immediately appointed to interim chief executive officer. Hours after Altman's removal, Brockman resigned as chairman, joined by director of research Jakub Pachocki and researchers Aleksander Mądry and Szymon Sidor. During an all-hands meeting, Sutskever defended the ouster and denied accusations of a hostile takeover. An OpenAI representative requested former board member Will Hurd's presence.

2. Related Work, Literature Survey & Algorithmic Foundations

🎥 Background Remover with U-Net + ResNet50 A deep learning project for automatic background removal from images and videos using a custom U-Net architecture with a ResNet50 encoder. The model performs semantic segmentation to identify and remove backgrounds, replacing them with a customizable color (default: magenta). 📋 Table of Contents Features Project Structure Model Architecture Requirements Installation Usage Training Image Prediction Video Processing Configuration Customization Troubleshooting License ✨ Features 🎯 High Accuracy : U-Net architecture with ResNet50 encoder for precise segmentation 🎬 Video Support : Process entire videos frame-by-frame with progress tracking 🖼️ Image Support : Remove backgrounds from individual images 🎨 Customizable Background : Replace removed backgrounds with any color 📊 Training Metrics : Track IoU (Intersection over Union) during training 💾 Model Checkpointing : Automatic saving of best model during training 📁 Project Structure . ├── dataset/ │ ├── images/ # Input training images (.png) │ └── masks/ # Corresponding binary masks (1 channel, .png) ├── files/ │ ├── model.h5 # Trained model checkpoint (best model) │ ├── best_model.h5 # Alternative model checkpoint │ └── data.csv # Training log (generated after training) ├── test/ │ ├── images/ # Test images for prediction │ └── masks/ # Generated prediction outputs ├── videos/ │ ├── inputs/ # Input videos (.mp4) │ └── outputs/ # Output processed videos (.avi) ├── model.py # U-Net + ResNet50 model architecture ├── train.py # Training script ├── predict.py # Background removal for images ├── videos.py # Background removal for videos ├── Background_remover.ipynb # Jupyter notebook (optional) └── README.md 🧠 Model Architecture Architecture Overview The model uses a U-Net architecture with the following components: Encoder : ResNet50 (pretrained on ImageNet) as the backbone Extracts multi-scale features from input images Uses skip connections at different resolution levels Decoder : Custom decoder with: Residual blocks : For better gradient flow and feature refinement Dilated convolutions : Multi-scale context aggregation (dilation rates: 3, 6, 9) Transposed convolutions : Upsampling with skip connections Multi-scale outputs : 4 different resolution predictions combined via concatenation Loss Function & Metrics Loss : Combined Binary Crossentropy + Dice Loss Binary Crossentropy: Standard pixel-wise classification loss Dice Loss: Handles class imbalance and improves segmentation boundaries Metric : IoU (Intersection over Union) Measures overlap between predicted and ground truth masks Higher IoU = better segmentation accuracy Training Configuration Input Size : 256×256 pixels Batch Size : 4 Learning Rate : 1e-4 (with ReduceLROnPlateau callback) Epochs : 100 (with early stopping) Validation Split : 20% Optimizer : Adam 📦 Requirements Python Version Python 3.7+ Dependencies tensorflow>=2.0.0 opencv-python>=4.5.0 numpy>=1.19.0 scikit-learn>=0.24.0 tqdm>=4.60.0 Hardware Recommendations GPU : NVIDIA GPU with CUDA support (recommended for faster training) RAM : 8GB+ recommended Storage : Sufficient space for dataset and model files 🚀 Installation 1. Clone or Download the Repository cd " Background remover " 2. Install Dependencies pip install tensorflow opencv-python tqdm numpy scikit-learn Note : For GPU support, install TensorFlow with GPU: pip install tensorflow-gpu 3. Verify Installation python -c " import tensorflow as tf; print(tf.__version__) " 💻 Usage 1. Training Prepare Your Dataset Organize your training data: dataset/ ├── images/ │ ├── image1.png │ ├── image2.png │ └── ... └── masks/ ├── image1.png ├── image2.png └── ... Dataset Requirements : Images and masks must have matching filenames Masks must be single-channel grayscale images Foreground (object): White pixels ( 255 ) Background: Black pixels ( 0 ) Supported format: PNG Run Training python train.py Training Process : Automatically splits data into train/validation (80/20) Saves best model to files/model.h5 based on validation loss Logs training metrics to files/data.csv Includes callbacks: ModelCheckpoint : Saves best model ReduceLROnPlateau : Reduces learning rate when plateau detected EarlyStopping : Stops training if no improvement for 15 epochs CSVLogger : Logs training history Expected Output : Epoch 1/100 ... Model saved to files/model.h5 ... 2. Image Prediction Remove backgrounds from individual images in the test/images/ directory. Prepare Test Images Place your test images in: test/images/ ├── your_image1.jpg ├── your_image2.png └── ... Supported formats: .jpg , .jpeg , .png , .webp Run Prediction python predict.py Output : Processed images saved to test/masks/ Each output shows: Original image | Separator | Masked image (background removed) Images are automatically resized to 256×256 for processing, then resized back to original dimensions Example : Images found: 12 Loading model... Model loaded. Processing: 100% | ████████████ | 12/12 [00: 15< 00:00, 1.25s/it] Saved transparent background image: messi ... Done ! 3. Video Processing Remove backgrounds from entire videos frame-by-frame. Prepare Input Video Place your video in: videos/inputs/video-1.mp4 Supported Formats : MP4, AVI, MOV (any format supported by OpenCV) Run Video Processing python videos.py Processing Details : Processes each frame at 256×256 resolution Upscales predictions back to original video dimensions Replaces background with magenta color Output saved as .avi format (MJPG codec) Output Location : videos/outputs/video-1.avi Note : To process a different video, edit the video_name variable in videos.py : video_name = "your-video-name" # Without extension ⚙️ Configuration Training Hyperparameters Edit train.py to modify: image_h = 256 # Image height image_w = 256 # Image width batch_size = 4 # Batch size (adjust based on GPU memory) lr = 1e-4 # Learning rate num_epochs = 100 # Maximum epochs Model Input Size The model expects 256×256 input. To change this: Modify image_h and image_w in train.py Update input_shape in build_model() call Retrain the model 🛠 Customization Change Background Color For Videos ( videos.py ) Edit line 73 in videos.py : background_mask = background_mask * [ 255 , 0 , 255 ] # Magenta (R, G, B) Examples : White: [255, 255, 255] Black: [0, 0, 0] Green: [0, 255, 0] Blue: [0, 0, 255] Custom: [R, G, B] (values 0-255) For Images ( predict.py ) Currently outputs masked images. To add background color: Modify predict.py after line 50 Add similar logic as in videos.py Transparent Background To output PNG with alpha channel (transparent background): Modify the output format in predict.py or videos.py Use 4-channel output (RGBA) instead of 3-channel (RGB) Set alpha channel based on mask values Video Frame Rate Edit line 46 in videos.py : out = cv2 . VideoWriter (..., 30 , ...) # Change 30 to desired FPS 🔧 Troubleshooting Common Issues 1. Out of Memory Error Solution : Reduce batch_size in train.py (try 2 or 1) Enable GPU memory growth (already included in code) 2. Model Not Found Error Solution : Ensure files/model.h5 exists. Train the model first using python train.py 3. No Images Found Solution : Check that images are in the correct directory: Training: dataset/images/ and dataset/masks/ Testing: test/images/ 4. Mask/Image Mismatch Solution : Ensure filenames match exactly (case-sensitive) Verify mask format: single-channel grayscale PNG 5. Poor Segmentation Results Solution : Train for more epochs Check mask quality (clear foreground/background separation) Increase dataset size Adjust learning rate or batch size 6. Video Processing is Slow Solution : Use GPU for inference Reduce video resolution before processing Process fewer frames per second GPU Setup If you have an NVIDIA GPU: Install CUDA and cuDNN Install TensorFlow GPU: pip install tensorflow-gpu Verify GPU detection: import tensorflow as tf print ( tf . config . list_physical_devices ( 'GPU' )) 📊 Training Tips Dataset Quality : Ensure high-quality masks with clear boundaries Data Augmentation : Consider adding augmentation for better generalization Learning Rate : Start with 1e-4, adjust if loss doesn't decrease Early Stopping : Monitor validation loss to prevent overfitting Model Checkpointing : Best model is automatically saved based on validation loss 📝 Notes The model outputs 4-channel masks during training (for multi-scale supervision), but uses the last channel for inference Input images are normalized to [0, 1] range Masks are resized to match input dimensions (256×256) during training Video output uses MJPG codec for compatibility 📄 License MIT License. Free to use, modify, and distribute. 🙏 Credits Model Architecture : U-Net with ResNet50 encoder Deep Learning Framework : TensorFlow/Keras Image Processing : OpenCV Progress Tracking : tqdm 🔗 Related Resources U-Net Paper ResNet Paper TensorFlow Documentation OpenCV Documentation

A deep learning project for automatic background removal from images and videos using a custom U-Net architecture with a ResNet50 encoder. The model performs semantic segmentation to identify and remove backgrounds, replacing them with a customizable color (default: magenta).

. ├── dataset/ │ ├── images/ # Input training images (.png) │ └── masks/ # Corresponding binary masks (1 channel, .png) ├── files/ │ ├── model.h5 # Trained model checkpoint (best model) │ ├── best_model.h5 # Alternative model checkpoint │ └── data.csv # Training log (generated after training) ├── test/ │ ├── images/ # Test images for prediction │ └── masks/ # Generated prediction outputs ├── videos/ │ ├── inputs/ # Input videos (.mp4) │ └── outputs/ # Output processed videos (.avi) ├── model.py # U-Net + ResNet50 model architecture ├── train.py # Training script ├── predict.py # Background removal for images ├── videos.py # Background removal for videos ├── Background_remover.ipynb # Jupyter notebook (optional) └── README.md

tensorflow>=2.0.0 opencv-python>=4.5.0 numpy>=1.19.0 scikit-learn>=0.24.0 tqdm>=4.60.0

dataset/ ├── images/ │ ├── image1.png │ ├── image2.png │ └── ... └── masks/ ├── image1.png ├── image2.png └── ...

Images found: 12 Loading model... Model loaded. Processing: 100% | ████████████ | 12/12 [00: 15< 00:00, 1.25s/it] Saved transparent background image: messi ... Done !

image_h = 256 # Image height image_w = 256 # Image width batch_size = 4 # Batch size (adjust based on GPU memory) lr = 1e-4 # Learning rate num_epochs = 100 # Maximum epochs

3. Dataset Ingestion, Preprocessing & Stochastic Augmentations

A deep learning project for **automatic background removal** from images and videos using a custom U-Net architecture with a ResNet50 encoder. The model performs semantic segmentation to identify and remove backgrounds, replacing them with a customizable color (default: magenta).

Deep Learning Background Removal Tool A deep learning-based background removal tool using U-Net for pixel-wise semantic segmentation. This project involved implementing the architecture in PyTorch on a custom dataset of 5,000+ images, predicting object masks to output images with transparent backgrounds. Model: U-Net (Convolutional Neural Network) for semantic segmentation. Objective: Automatically remove the background from an image and make it transparent. Key Performance: Achieved a mean Intersection over Union (IoU) score of 0.89 on validation data. Achieved a pixel-wise accuracy of 96%. The trained model is packaged as a reusable Python module for easy integration. Input: RGB images (JPEG/PNG format). Output: RGBA images with transparent background (PNG format). Dataset: Trained on the DUTS-TE dataset (5,000+ images) and corresponding binary masks. Download: DUTS Saliency Detection Dataset Preprocessing: Applied extensive augmentation (flips, rotations, color jitter) to enhance robustness against diverse backgrounds. Frameworks: PyTorch, OpenCV, NumPy, PIL. Training: Model trained using binary cross-entropy (BCE) loss. Prediction: Single image or batch input supported; output saved with transparency.

A deep learning-based background removal tool using U-Net for pixel-wise semantic segmentation. This project involved implementing the architecture in PyTorch on a custom dataset of 5,000+ images, predicting object masks to output images with transparent backgrounds.

Dataset: Trained on the DUTS-TE dataset (5,000+ images) and corresponding binary masks.

Preprocessing: Applied extensive augmentation (flips, rotations, color jitter) to enhance robustness against diverse backgrounds.

A deep learning-based background removal tool using U-Net for image segmentation. Trained on paired images and masks, it predicts object masks and outputs images with transparent backgrounds

🖼️ AI Background Remove A lightweight yet powerful AI-based background removal tool built with Python. Automatically detects and removes backgrounds from images using deep learning — no manual masking required. Installation • Usage • How It Works • Example • Contributing 📌 Table of Contents Overview Features How It Works Project Structure Tech Stack Installation Usage Example Output Use Cases Troubleshooting Contributing License 🧠 Overview AI-BackgroundRemove is a Python script that leverages state-of-the-art deep learning to remove backgrounds from images in seconds. Powered by U²-Net (via the rembg library), it produces clean, high-quality transparent PNG outputs — perfect for e-commerce, design, and automation workflows. No Photoshop. No manual work. Just AI. ✨ Features 🤖 AI-powered — Uses U²-Net deep learning model for accurate segmentation 🖼️ Transparent PNG output — Removes background, exports with alpha channel ⚡ Fast & lightweight — Single Python script, minimal setup 📁 Single image & batch processing support 🎯 High accuracy — Works on people, products, animals, and objects 💻 Cross-platform — Windows, macOS, Linux ⚙️ How It Works Input Image (JPG / PNG / WEBP) │ ▼ ┌─────────────────────────────┐ │ rembg Library │ │ │ │ ┌─────────────────────┐ │ │ │ U²-Net Model │ │ ← Deep learning segmentation model │ │ (Salient Object │ │ trained on DUTS-TR dataset │ │ Detection) │ │ │ └──────────┬──────────┘ │ │ │ │ │ ┌──────────▼──────────┐ │ │ │ Alpha Matte │ │ ← Generates pixel-level foreground mask │ │ Generation │ │ │ └──────────┬──────────┘ │ └─────────────┼───────────────┘ │ ▼ ┌─────────────────────────────┐ │ Pillow (PIL) │ ← Applies mask, saves transparent PNG └─────────────────────────────┘ │ ▼ Output Image (PNG with transparency) U²-Net Architecture U²-Net uses a nested U-structure with two levels: Outer U-Net — Captures full-resolution spatial context Inner U-Net — Learns fine-grained local features inside each block This dual-nested design achieves high accuracy on salient object detection without requiring a pre-trained backbone. 🗂️ Project Structure AI-BackgroundRemove/ │ ├── 🐍 remove.py # Core script — loads image, removes bg, saves output ├── 📦 requirements.txt # Python dependencies ├── 📄 LICENSE # Apache 2.0 └── 📄 README.md # You are here 🛠️ Tech Stack Library Version Purpose rembg latest AI background removal via U²-Net Pillow ≥ 9.0 Image loading, manipulation & saving onnxruntime ≥ 1.13 ONNX model inference (used by rembg) numpy ≥ 1.21 Array operations on pixel data 📦 Installation 1. Clone the repository: git clone https://github.com/eddiebrock911/AI-BackgroundRemove.git cd AI-BackgroundRemove 2. Create & activate a virtual environment: # Create python -m venv venv # Activate — Linux/Mac source venv/bin/activate # Activate — Windows venv \S cripts \a ctivate 3. Install dependencies: pip install -r requirements.txt On first run, rembg will automatically download the U²-Net model weights (~170MB). Ensure you have an internet connection. ▶️ Usage Basic — Remove background from a single image python remove.py In your own Python code from rembg import remove from PIL import Image import io # Open input image input_image = Image . open ( "input.jpg" ) # Remove background output_image = remove ( input_image ) # Save as transparent PNG output_image . save ( "output.png" ) print ( "✅ Background removed successfully!" ) Batch Processing (multiple images) import os from rembg import remove from PIL import Image input_folder = "images/input/" output_folder = "images/output/" os . makedirs ( output_folder , exist_ok = True ) for filename in os . listdir ( input_folder ): if filename . lower (). endswith (( ".jpg" , ".jpeg" , ".png" , ".webp" )): img = Image . open ( os . path . join ( input_folder , filename )) result = remove ( img ) result . save ( os . path . join ( output_folder , f" { os . path . splitext ( filename )[ 0 ] } .png" )) print ( f"✅ Processed: { filename } " ) 📋 Example Output Input Output input.jpg (with background) output.png (transparent background) Person standing in a park Person isolated on transparent PNG Product on white table Product with pure transparent background Animal in nature Animal cleanly segmented Supported input formats: .jpg , .jpeg , .png , .webp , .bmp Output format: .png (with alpha transparency channel) 💡 Use Cases Domain Application 🛒 E-commerce Product photo cleanup for listings 🎨 Graphic Design Extract subjects for compositing 📸 Photography Portrait background replacement 🤖 ML Pipelines Preprocessing images for training datasets 🪪 ID / Passport Photos Auto white background generation 📱 Social Media Profile picture background removal 🐛 Troubleshooting Problem Solution ModuleNotFoundError: rembg Run pip install -r requirements.txt Model download fails Check internet connection; rembg downloads model on first run Output image is fully transparent Ensure input image is not corrupted or too small Slow processing Install onnxruntime-gpu for CUDA GPU acceleration PIL import error Run pip install Pillow Enable GPU Acceleration (Optional) pip uninstall onnxruntime pip install onnxruntime-gpu Requires CUDA-compatible NVIDIA GPU and CUDA toolkit installed. 🤝 Contributing Contributions are welcome! # 1. Fork the repo on GitHub # 2. Clone your fork git clone https://github.com/your-username/AI-BackgroundRemove.git # 3. Create a feature branch git checkout -b feature/your-feature-name # 4. Make your changes & commit git commit -m " feat: describe your change " # 5. Push & open a Pull Request git push origin feature/your-feature-name Ideas for contributions: 🖥️ Add a Streamlit / Gradio web UI 🗃️ Add CLI argument support ( --input , --output , --batch ) 🎨 Custom background color/image replacement 📊 Add image quality metrics 📄 License This project is licensed under the Apache 2.0 License — see the LICENSE file for details. Made with ❤️ by eddiebrock911 ⭐ Star this repo if it saved you time!

4. Neural Network Backbone, Layer Specifications & Loss Formulations

A lightweight yet powerful AI-based background removal tool built with Python. Automatically detects and removes backgrounds from images using deep learning — no manual masking required.

AI-BackgroundRemove is a Python script that leverages state-of-the-art deep learning to remove backgrounds from images in seconds. Powered by U²-Net (via the rembg library), it produces clean, high-quality transparent PNG outputs — perfect for e-commerce, design, and automation workflows.

Input Image (JPG / PNG / WEBP) │ ▼ ┌─────────────────────────────┐ │ rembg Library │ │ │ │ ┌─────────────────────┐ │ │ │ U²-Net Model │ │ ← Deep learning segmentation model │ │ (Salient Object │ │ trained on DUTS-TR dataset │ │ Detection) │ │ │ └──────────┬──────────┘ │ │ │ │ │ ┌──────────▼──────────┐ │ │ │ Alpha Matte │ │ ← Generates pixel-level foreground mask │ │ Generation │ │ │ └──────────┬──────────┘ │ └─────────────┼───────────────┘ │ ▼ ┌─────────────────────────────┐ │ Pillow (PIL) │ ← Applies mask, saves transparent PNG └─────────────────────────────┘ │ ▼ Output Image (PNG with transparency)

This dual-nested design achieves high accuracy on salient object detection without requiring a pre-trained backbone.

AI-BackgroundRemove/ │ ├── 🐍 remove.py # Core script — loads image, removes bg, saves output ├── 📦 requirements.txt # Python dependencies ├── 📄 LICENSE # Apache 2.0 └── 📄 README.md # You are here

git clone https://github.com/eddiebrock911/AI-BackgroundRemove.git cd AI-BackgroundRemove

# Create python -m venv venv # Activate — Linux/Mac source venv/bin/activate # Activate — Windows venv \S cripts \a ctivate

5. Training Protocol, Convergence Behavior & Experimental Evaluation

On first run, rembg will automatically download the U²-Net model weights (~170MB). Ensure you have an internet connection.

from rembg import remove from PIL import Image import io # Open input image input_image = Image . open ( "input.jpg" ) # Remove background output_image = remove ( input_image ) # Save as transparent PNG output_image . save ( "output.png" ) print ( "✅ Background removed successfully!" )

import os from rembg import remove from PIL import Image input_folder = "images/input/" output_folder = "images/output/" os . makedirs ( output_folder , exist_ok = True ) for filename in os . listdir ( input_folder ): if filename . lower (). endswith (( ".jpg" , ".jpeg" , ".png" , ".webp" )): img = Image . open ( os . path . join ( input_folder , filename )) result = remove ( img ) result . save ( os . path . join ( output_folder , f" { os . path . splitext ( filename )[ 0 ] } .png" )) print ( f"✅ Processed: { filename } " )

Supported input formats: .jpg , .jpeg , .png , .webp , .bmp Output format: .png (with alpha transparency channel)

# 1. Fork the repo on GitHub # 2. Clone your fork git clone https://github.com/your-username/AI-BackgroundRemove.git # 3. Create a feature branch git checkout -b feature/your-feature-name # 4. Make your changes & commit git commit -m " feat: describe your change " # 5. Push & open a Pull Request git push origin feature/your-feature-name

This project is licensed under the Apache 2.0 License — see the LICENSE file for details.

A lightweight yet powerful AI-based background removal tool built with Python. Automatically detects and removes backgrounds from images using deep learning — no manual masking required.