| --- |
| license: mit |
| language: |
| - en |
| tags: |
| - image-classification |
| - multi-label |
| - resnet |
| - pytorch |
| --- |
| # Multi-Label Object Classification using ResNet |
|
|
| ## Model Description |
| ResNet18 and ResNet50 models fine-tuned for multi-label object classification, |
| capable of detecting 12 objects simultaneously in a single image. |
|
|
| ## Classes |
| `backpack`, `book`, `bottle`, `calculator`, `chair`, `clock`, |
| `desk`, `keychain`, `laptop`, `paper`, `pen`, `phone` |
|
|
| ## Usage |
| Download all 10 `.pth` files and use with the inference script from the |
| [GitHub repository](https://github.com/pranav1233/multi-label-object-classification). |
|
|
| ## Model Files |
| - `resnet18_fold1.pth` through `resnet18_fold5.pth` — ResNet18 ensemble |
| - `resnet50_fold1.pth` through `resnet50_fold5.pth` — ResNet50 ensemble |
|
|
| ## Performance |
|
|
| | Model | Exact Match | Micro F1 | Mean IOU | |
| |----------|-------------|----------|----------| |
| | ResNet18 | 52.29% | 76.92% | 0.7200 | |
| | ResNet50 | 68.78% | 86.18% | 0.8301 | |
|
|
| ## Training |
| - 5-Fold Stratified Cross Validation |
| - Test Time Augmentation (TTA) with 10 augmented views |
| - Optimal prediction threshold: 0.40 |
