TRAM-V1

Reference checkpoint for TRAM-V1 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 + Mask R-CNN
• Input: 224×224 RGB
• Training data: SSS_OD-5 (SeabedObjects-KLSG-derived, plane + ship)
• Best epoch: 19 / 100
• Random seed: not fixed

Validation metrics (this run)

Metric	This run	Paper
Det mAP@0.5	0.8926	0.8657
Det mAP@0.5:0.95	0.5588	0.5391
Seg mAP@0.5	0.7925	0.7226
Seg mAP@0.5:0.95	0.3945	0.4098

Usage

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_v1.inference \
    --weights /path/to/tram_v1_best.pth \
    --data-root /path/to/SSS_OD-5/valid \
    --output-dir runs/tram-v1/inference \
    --evaluate-map

Files

• tram_v1_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 · tram-v2 · tram-v3