| # 3D Object Dimension & Volume Estimator |
|
|
| This project is a Streamlit web application that estimates the 3D dimensions (Length, Width, Height) and Volume of objects from user-uploaded images. It leverages Detectron2 for object detection and instance segmentation, and a custom Convolutional Neural Network (CNN) trained on the Pix3D dataset for dimension regression. |
|
|
| ## Features |
|
|
| * Upload single or multiple images of an object (different views). |
| * Detects objects using a pre-trained Detectron2 (Mask R-CNN) model. |
| * Displays segmentation masks and 2D bounding boxes for detected objects. |
| * For the largest detected object in each view: |
| * Crops the object using its segmentation mask. |
| * Feeds the cropped patch to a custom CNN to predict dimensions (L, W, H, V). |
| * Displays individual dimension predictions for each view. |
| * Calculates and displays aggregated (averaged) dimensions if multiple views are provided. |
| * User-friendly web interface built with Streamlit. |
|
|
| ## Models Used |
|
|
| 1. **Object Detection & Segmentation:** |
| * **Detectron2 (Mask R-CNN R50-FPN 3x):** Pre-trained on the COCO dataset. Used to identify objects and generate pixel-wise segmentation masks. |
| 2. **Dimension Estimation:** |
| * **Custom CNN (ResNet50 backbone):** Trained on image patches derived from the Pix3D dataset. The model takes a cropped image patch of an object as input and outputs its estimated Length, Width, Height (in meters), and Volume (in meters³). |
|
|
| ## Setup and Installation |
|
|
| Follow these steps to set up and run the application locally: |
|
|
| **1. Clone this GitHub Repository:** |
|
|
| ```bash |
| git clone https://huggingface.co/suryaprakash01/dimension_Detect/edit/main |
| cd YourGitHubRepoName |
