---
license: cc-by-4.0
tags:
  - object-detection
  - instance-segmentation
  - mask-rcnn
  - swin-transformer
  - sonar
  - underwater
library_name: pytorch
pipeline_tag: object-detection
---

# TRAM-V2

Reference checkpoint for **TRAM-V2** from the paper *TRAM:
Transformer-Based Mask R-CNN Framework for Underwater Object Detection in
Side-Scan Sonar Data* (Makam, Sundaram, & Sundaram).

- **Architecture**: MST (Swin-Base) + FPN + CBAM + Mask R-CNN
- **Input**: 224×224 RGB
- **Training data**: SSS_OD-5 (SeabedObjects-KLSG-derived, plane + ship)
- **Best epoch**: 27 / 47
- **Random seed**: not fixed

## Validation metrics (this run)

| Metric | This run | Paper |
|---|---|---|
| Det mAP@0.5 | 0.8356 | 0.8321 |
| Det mAP@0.5:0.95 | 0.5496 | 0.5293 |
| Seg mAP@0.5 | 0.7538 | 0.7352 |
| Seg mAP@0.5:0.95 | 0.4244 | 0.4412 |

## Usage

```bash
git clone -b final-tram-v123 https://github.com/CrypticCortex/iisc-sss-codes.git
cd iisc-sss-codes
pip install -r final/requirements.txt

python -m final.tram_v2.inference \
    --weights /path/to/tram_v2_best.pth \
    --data-root /path/to/SSS_OD-5/valid \
    --output-dir runs/tram-v2/inference \
    --evaluate-map
```

## Files

- `tram_v2_best.pth` — best checkpoint (highest val bbox mAP@0.5:0.95)
- `training.log` — per-epoch training/validation log


## Source

- Code: https://github.com/CrypticCortex/iisc-sss-codes/tree/final-tram-v123/final
- Other variants: [tram-v1](https://huggingface.co/aghoraguru/tram-v1) · [tram-v2](https://huggingface.co/aghoraguru/tram-v2) · [tram-v3](https://huggingface.co/aghoraguru/tram-v3)