---
license: other
license_name: research-only
tags:
  - image-compression
  - video-diffusion
  - ultra-low-bitrate
---

<img src="./assets/logo_nefic_true.png" style="zoom: 33%;" />

# NeFIC: Next-Frame Decoding for Ultra-Low-Bitrate Image Compression with Video Diffusion Priors

**ECCV 2026**

**Authors:** Yunuo Chen, Chuqin Zhou, Jiangchuan Li, Xiaoyue Ling, Bing He, Jincheng Dai, Li Song, Guo Lu

## Links

- [📄 Paper (arXiv)](https://arxiv.org/abs/2603.15129)
- [💻 Code (GitHub)](https://github.com/UnoC-727/NeFIC)

## Checkpoints

This repository contains 6 pretrained NeFIC checkpoints at different rate points:

| Directory | Lambda | Rate |
|-----------|--------|------|
| `checkpoints/nefic_lambda0.25/` | 0.25 | Highest bitrate |
| `checkpoints/nefic_lambda0.5/` | 0.5 | |
| `checkpoints/nefic_lambda1/` | 1.0 | |
| `checkpoints/nefic_lambda1.7/` | 1.7 | |
| `checkpoints/nefic_lambda3/` | 3.0 | |
| `checkpoints/nefic_lambda5/` | 5.0 | Lowest bitrate |

Each directory contains:
- `anchor_codec.bin` — Anchor Codec weights
- `pytorch_lora_weights.safetensors` — LoRA adapter for the generative decoder

## Usage

```bash
hf download yuUnuo/NeFIC --local-dir /path/to/nefic_checkpoints
```

See the [GitHub repo](https://github.com/UnoC-727/NeFIC) for full inference instructions.