---
license: other
tags:
  - watermarking
  - image-watermarking
  - adversarial-training
  - robust-watermarking
---

# CAT-Pixelseal-Compositional

This is the **CAT (Compositional Adversarial Training)** checkpoint for PixelSeal with a **compositional adversary (T=2)**.

## Overview

CAT is a plug-in training framework that replaces random augmentation with a learned sequential adversary to improve watermark robustness. Instead of sampling augmentations uniformly and independently, CAT trains a lightweight adversary—a frozen DINOv2 backbone + GRU controller + MLP heads—that adaptively selects a sequence of augmentations most likely to break the current watermark model at each training step.

This checkpoint corresponds to the **T=2 (compositional)** setting with:
- **Param head type**: `random_uniform` — augmentation parameters are sampled uniformly at random within learned ranges
- **Adversary depth**: 2 (two adversarial augmentations chained per training step)
- **Entropy weight** (λ_ent): 0.1 — entropy regularization to prevent collapse to a single attack
- **VI alpha** (α): 0.0

## Architecture

| Component | Config |
|---|---|
| Embedder | `unet_small2_yuv_quant` — U-Net operating in YUV space, 8 blocks, batch norm |
| Extractor | `convnext_tiny` — ConvNeXt-Tiny encoder + pixel decoder |
| Payload | 16 bits |
| Adversary | DINOv2 backbone + GRU + MLP, hidden dim 256, Gumbel-Softmax τ=1.0 |

## Training

Trained with CAT (T=2) on SA-1B using 4 GPUs:

```bash
OMP_NUM_THREADS=40 torchrun --nproc_per_node=4 train.py --local_rank 0 \
    --video_dataset none --image_dataset sa-1b-full-resized --workers 4 \
    --extractor_model convnext_tiny --embedder_model unet_small2_yuv_quant \
    --hidden_size_multiplier 1 --nbits 16 \
    --scaling_w_schedule Cosine,scaling_min=0.2,start_epoch=200,epochs=200 \
    --scaling_w 1.0 --scaling_i 1.0 --attenuation jnd_1_1 \
    --epochs 500 --iter_per_epoch 1000 \
    --scheduler CosineLRScheduler,lr_min=1e-6,t_initial=500,warmup_lr_init=1e-8,warmup_t=5 \
    --optimizer AdamW,lr=5e-4 \
    --lambda_dec 1.0 --lambda_d 0.1 --lambda_i 0.1 --perceptual_loss yuv \
    --num_augs 2 --augmentation_config configs/all_augs.yaml \
    --disc_in_channels 1 --disc_start 50 \
    --use_adversary True \
    --adversary_entropy_weight 0.1 \
    --adversary_hidden_dim 256 \
    --adversary_gumbel_temperature 1.0 \
    --adversary_param_head_type random_uniform \
    --adversary_vi_alpha 0.0 \
    --adversary_start_epoch 5 \
    --percep_loss_start_epoch 50
```

## Augmentation Library

Trained against 15 augmentation families (see `configs/augs.yaml`): identity, JPEG, crop, rotate, rotate90, horizontal flip, perspective, Gaussian blur, brightness, contrast, saturation, hue, H.264, H.264 RGB, H.265.

## Usage

See the [CAT repository](https://github.com/facebookresearch/videoseal) and `configs/` for architecture details. Load with:

```python
from videoseal.models.videoseal import Videoseal
model = Videoseal.load_from_checkpoint("adv_d2_phtrandom_uniform_a0.0_ew0.10/checkpoint.pth")
```

## Citation

```bibtex
@article{cat2025,
  title={Compositional Adversarial Training for Robust Visual Watermarking},
  author={Anonymous Authors},
  year={2025},
}
```

## License

Inherits the license from the VideoSeal repository (Meta Platforms, Inc.).