🚮 Trashify Object Detection Demo

streamlit link (hosted) : https://ry438wpt4hzqbn4app33cux.streamlit.app/

Trashify is an object detection application aimed at helping clean up local areas! Upload an image, and the system evaluates if it contains a trash, bin, and hand. If all three are detected, you receive a +1!

🚀 Features

• Gamified Object Detection: Earn points by proving you picked up trash and disposed of it correctly.
• State-of-the-Art Model: Powered by a fine-tuned version of RT-DETRv2 trained on the Trashify dataset.
• Interactive UI: Built using Streamlit, featuring an intuitive drag-and-drop interface, confidence threshold sliders, and real-time inference with bounding box annotations.
• Model Hosting: The model is hosted on Hugging Face: RahulKate-173/rt_detrv2_finetuned_trashify_box_detector_v2.

📁 Repository Structure

• app.py: The main Streamlit application script containing the UI layout, inference pipeline, and the gamification rule engine.
• notebooks/: A collection of Jupyter Notebooks detailing the end-to-end process of building Trashify:
  • Trashify_2_model.ipynb: Loading the dataset, preprocessing in COCO format, training/fine-tuning the RT-DETRv2 model, and pushing to Hugging Face.
  • Building_trashify_with_augmentation.ipynb: Experiments utilizing data augmentation.
  • Non_Mamixum_Supperission (1).ipynb: Deep dives into object detection concepts like Non-Maximum Suppression (NMS).
  • Other notebooks detailing the iterative process of building the model and interface.
• requirements.txt: The required Python packages to run the application and notebooks.

🛠️ How to Run Locally

1. Clone the repository:

   git clone <repo-url>
   cd trashify

2. Install the dependencies: Ensure you have Python installed. It is recommended to use a virtual environment.

   pip install -r requirements.txt

   (Note: Core dependencies include torch, streamlit, Pillow, and transformers).

3. Run the Streamlit application:

   streamlit run app.py

4. Open in Browser: The application should automatically open in your default browser at http://localhost:8501.

🧠 Model Classes & Colors

The model detects the following classes, which are annotated with specific colors during inference:

• Bin: Green (#00e676)
• Trash: Blue (#2979ff)
• Hand: Purple (#aa00ff)
• Trash Arm: Yellow (#ffd600)
• Not Trash / Not Bin / Not Hand: Red (#ff1744)