VAD-JEPA Model Checkpoints
Pretrained temporal model weights for the VAD-JEPA project — online video anomaly detection on traffic dashcam footage.
Variants
| Model | VCL | NF | Type | Best Epoch |
|---|---|---|---|---|
| SG-SlotSSM (Sparse Gated) | 64 | 4 | frozen | 40 |
| SG-SlotSSM (Sparse Gated k = 2) | 64 | 4 | finetuned | 20 |
| SG-SlotSSM (Sparse Gated k = 4) | 64 | 4 | finetuned | 20 |
| SG-SlotSSM (Sparse Gated k = 8) | 64 | 4 | finetuned | 20 |
| SG-SlotSSM (Sparse Gated) | 64 | 4 | finetuned | 20 |
| SG-SlotSSM (Sparse Gated) | 28 | 4 | frozen | 90 |
| SG-SlotSSM (Sparse Gated) | 28 | 4 | finetuned | 60 |
| SG-SlotSSM (Sparse Gated) | 8 | 4 | frozen | 100 |
| SG-SlotSSM (Sparse Gated) | 8 | 4 | finetuned | 90 |
| SlotSSM (Dense) | 64 | 4 | frozen | 30 |
| SlotSSM (Dense) | 64 | 4 | finetuned | 30 |
| SlotSSM (Dense) | 28 | 4 | frozen | 100 |
| SlotSSM (Dense) | 28 | 4 | finetuned | 90 |
| SlotSSM (Dense) | 8 | 4 | frozen | 100 |
| SlotSSM (Dense) | 8 | 4 | finetuned | 100 |
| Mamba | 64 | 4 | frozen | 40 |
| Mamba | 64 | 4 | finetuned | 20 |
| Mamba | 28 | 4 | frozen | 60 |
| Mamba | 28 | 4 | finetuned | 30 |
| Mamba | 8 | 4 | frozen | 90 |
| Mamba | 8 | 4 | finetuned | 100 |
| LSTM | 64 | 4 | frozen | 20 |
| LSTM | 64 | 4 | finetuned | 20 |
| LSTM | 28 | 4 | frozen | 50 |
| LSTM | 28 | 4 | finetuned | 40 |
| LSTM | 8 | 4 | frozen | 100 |
| LSTM | 8 | 4 | finetuned | 80 |
| Encoder Only | 8 | 4 | finetuned | 150 |
Usage
Download a checkpoint and use it with the VAD-JEPA repo:
python main.py --config cfgs/vjepa_sparse_slotssm.yaml --phase test --epoch 50
Each folder contains checkpoints/model-{epoch}.pt.