PixelREPA - GitLite

KO | EN

gitlite — search

#javascript #python #react #go #golang #machine-learning #docker #macos #nodejs #java #kubernetes #awesome-list #mcp #openai

PixelREPA

(★ 52)

[ECCV 2026] Official code of "Representation Alignment for Just Image Transformers is not Easier than You Think"

File Explorer

Download Latest Version (.zip)

- Fig1.png
- framework.jpg
- PixelREPA.pdf
- jit_in256_stats.npz
- crop.py
- lr_sched.py
- misc.py
- model_util.py
.gitignore
denoiser_pixelREPA.py
engine_pixelREPA.py
environment.yaml
LICENSE
main_pixelREPA.py
model_pixelREPA.py
prepare_ref.py
README.md

# Installation Guide

1. Get the code

git clone https://github.com/kaist-cvml/PixelREPA

Downloads the entire project code from GitHub to your computer.

cd PixelREPA

Moves into the project folder you just downloaded.

2. Official Install Script

Easy Recommended

Prerequisites

Python 3 Python is required to use pip.

pip install torch==2.5.1 --index-url https://download.pytorch.org/whl/cu124

Installs the package published on PyPI directly — no need to clone the source.

✅ After installing, open a new terminal and run the program's version command (e.g. --version) to confirm it worked.

Pulled directly from this repo's README.

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# [ECCV 2026] PixelREPA: Representation Alignment for Just Image Transformers is not Easier than You Think

<div align="center">
  <a href="https://scholar.google.co.kr/citations?user=UbZM7nQAAAAJ&hl=ko&oi=ao" target="_blank">Jaeyo&nbsp;Shin</a><sup>*</sup> &ensp; <b>&middot;</b> &ensp;
  <a href="https://jw-kim.netlify.app/" target="_blank">Jiwook&nbsp;Kim</a><sup>*</sup> &ensp; <b>&middot;</b> &ensp;
  <a href="https://scholar.google.co.kr/citations?user=KB5XZGIAAAAJ&hl=ko" target="_blank">Hyunjung&nbsp;Shim</a> <br>
  KAIST AI &emsp;<br>
  <sup>*</sup>Equal Contribution &emsp; <br>
</div>
<br/>

<p align="center">
  <a href="https://arxiv.org/abs/2603.14366">
    <img src="https://img.shields.io/badge/arXiv%20paper-2603.14366-b31b1b.svg">
  </a>
  <a href="https://raw.githubusercontent.com/kaist-cvml/PixelREPA/main/demo/PixelREPA.pdf">
    <img src="https://img.shields.io/badge/PDF-View-blue">
  </a>
</p>

This repository is a PyTorch official code of the paper [Representation Alignment for Just Image Transformers is not Easier than You Think](https://arxiv.org/abs/2603.14366).

## News & Updates
  - [x] **[2026/06/18]** ✈️ Our paper is accepted to ECCV 2026!
  - [x] **[2026/03/17]** 📝 Our paper is now available! The paper is released [here](https://arxiv.org/abs/2603.14366).
  - [x] **[2026/03/16]** 🧑‍💻 Our code is released!

## PixelREPA

<p align="center">
  <img src="demo/Fig1.png" width="100%">
</p>

PixelREPA aligns JiT intermediate features to the feature space of an external semantic encoder using our Masked Transformer Adapter (MTA), instead of an MLP.
This alternative design accelerates training speed of JiT.
The detailed design is described as a following figure:
<br />

<p align="center">
  <img src="demo/framework.jpg" width="100%">
</p>

## 1. Environment setting and Dataset Preparation

### Code Download

```bash
git clone https://github.com/kaist-cvml/PixelREPA.git
cd PixelREPA
```

### Dataset Download

Download [ImageNet-10k](https://image-net.org/download).

### Conda Environment
```bash
conda env create -f environment.yaml
conda activate pixelrepa
```

If you get ```undefined symbol: iJIT_NotifyEvent``` when importing ```torch```, simply
```bash
pip uninstall torch
pip install torch==2.5.1 --index-url https://download.pytorch.org/whl/cu124
```
Check this [issue](https://github.com/conda/conda/issues/13812#issuecomment-2071445372) for more details.

## 2. Training

We trained on 8 NVIDIA H200 GPUs.

Here is the training script on ImageNet 256x256 for 600 epochs:

1. PixelREPA-B/16
```bash
  torchrun --nproc_per_node=8 --nnodes=1 --node_rank=0 \
    main_pixelREPA.py \
    --model JiT-B/16 \
    --mta MTA-B/16 \
    --proj_dropout 0.0 \
    --P_mean -0.8 --P_std 0.8 \
    --img_size 256 --noise_scale 1.0 \
    --batch_size 128 --blr 5e-5 \
    --epochs 600 --warmup_epochs 5 \
    --gen_bsz 128 --num_images 50000 --cfg 3.1 --interval_min 0.1 --interval_max 1.0 \
    --output_dir ${OUTPUT_DIR} --resume ${OUTPUT_DIR} \
    --data_path ${IMAGENET_PATH} --online_eval --eval_freq 10
```

2. PixelREPA-L/16
```bash
  torchrun --nproc_per_node=8 --nnodes=1 --node_rank=0 \
    main_pixelREPA.py \
    --model JiT-L/16 \
    --mta MTA-L/16 \
    --proj_dropout 0.0 \
    --P_mean -0.8 --P_std 0.8 \
    --img_size 256 --noise_scale 1.0 \
    --batch_size 128 --blr 5e-5 \
    --epochs 600 --warmup_epochs 5 \
    --gen_bsz 128 --num_images 50000 --cfg 2.8 --interval_min 0.1 --interval_max 1.0 \
    --output_dir ${OUTPUT_DIR} --resume ${OUTPUT_DIR} \
    --data_path ${IMAGENET_PATH} --online_eval --eval_freq 10
```

3. PixelREPA-H/16
```bash
  torchrun --nproc_per_node=8 --nnodes=1 --node_rank=0 \
    main_pixelREPA.py \
    --model JiT-H/16 \
    --mta MTA-H/16 \
    --proj_dropout 0.2 \
    --P_mean -0.8 --P_std 0.8 \
    --img_size 256 --noise_scale 1.0 \
    --batch_size 128 --blr 5e-5 \
    --epochs 600 --warmup_epochs 5 \
    --gen_bsz 128 --num_images 50000 --cfg 2.4 --interval_min 0.1 --interval_max 1.0 \
    --output_dir ${OUTPUT_DIR} --resume ${OUTPUT_DIR} \
    --data_path ${IMAGENET_PATH} --online_eval --eval_freq 10
```

## 3. Inference

Pre-trained models are available [here](https://drive.google.com/drive/folders/1TzFxFcqKzRfwkNo6AvndROMlIr8GmM3i?usp=sharing).

1. PixelREPA-B/16
```bash
  torchrun --nproc_per_node=8 --nnodes=1 --node_rank=0 \
    main_pixelREPA.py \
    --model JiT-B/16 \
    --mta MTA-B/16 \
    --img_size 256 --noise_scale 1.0 \
    --gen_bsz 128 --num_images 50000 --cfg 3.1 --interval_min 0.1 --interval_max 1.0 \
    --output_dir ${CKPT_DIR} --resume ${CKPT_DIR} \
    --data_path ${IMAGENET_PATH} --evaluate_gen
```

2. PixelREPA-L/16
```bash
  torchrun --nproc_per_node=8 --nnodes=1 --node_rank=0 \
    main_pixelREPA.py \
    --model JiT-L/16 \
    --mta MTA-L/16 \
    --img_size 256 --noise_scale 1.0 \
    --gen_bsz 128 --num_images 50000 --cfg 2.8 --interval_min 0.1 --interval_max 1.0 \
    --output_dir ${CKPT_DIR} --resume ${CKPT_DIR} \
    --data_path ${IMAGENET_PATH} --evaluate_gen
```

3. PixelREPA-H/16
```bash
  torchrun --nproc_per_node=8 --nnodes=1 --node_rank=0 \
    main_pixelREPA.py \
    --model JiT-H/16 \
    --mta MTA-H/16 \
    --img_size 256 --noise_scale 1.0 \
    --gen_bsz 128 --num_images 50000 --cfg 2.4 --interval_min 0.1 --interval_max 1.0 \
    --output_dir ${CKPT_DIR} --resume ${CKPT_DIR} \
    --data_path ${IMAGENET_PATH} --evaluate_gen
```

## Acknowledgement

Our code is strongly based on [JiT](https://github.com/LTH14/JiT). We sincerely appreciate to the following works:
  - [JiT](https://github.com/LTH14/JiT)
  - [REPA](https://github.com/sihyun-yu/REPA)

## Citation

```bibtex
@misc{shin2026pixelrepa,
      title={Representation Alignment for Just Image Transformers is not Easier than You Think}, 
      author={Jaeyo Shin and Jiwook Kim and Hyunjung Shim},
      year={2026},
      eprint={2603.14366},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2603.14366}, 
}
```