# Image Reconstruction on ImageNet 50k 1k (val)

Source: [Image Reconstruction on ImageNet 50k 1k (val)](<https://www.sota2.com/research/sota/image-reconstruction-on-imagenet-50k-1k-val>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-02-26T06:11:16.772Z

Evaluation of image reconstruction models using the ImageNet 50k 1k validation split, typically measured using metrics like rFID, PSNR, and SSIM.

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Image Reconstruction |
| Dataset (including any stated split) | ImageNet 50k 1k (val) |
| Metric | rFID |
| Unit | SCALAR |
| Direction | Lower is better |
| Results loaded | 25 of 25 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [rFID](<https://www.sota2.com/research/sota/image-reconstruction-on-imagenet-50k-1k-val.md?metric=rFID>): Lower is better; unit: SCALAR.

- [PSNR](<https://www.sota2.com/research/sota/image-reconstruction-on-imagenet-50k-1k-val.md?metric=PSNR>): Higher is better; unit: SCALAR.

- [SSIM](<https://www.sota2.com/research/sota/image-reconstruction-on-imagenet-50k-1k-val.md?metric=SSIM>): Higher is better; unit: PERCENT.

- [LPIPS](<https://www.sota2.com/research/sota/image-reconstruction-on-imagenet-50k-1k-val.md?metric=LPIPS>): Lower is better; unit: SCALAR.

## Recorded results

| Method | rFID | Result date | Report paper |
| --- | --- | --- | --- |
| TokenFlow | 0.63 | Not recorded | 68fa7a741e209131cea835a3 |
| Open-MAGVIT2-I-PT | 0.78 | Not recorded | 68fa76e01e209131cea6f87d |
| LlamaGen | 0.94 | Not recorded | 68fa7a741e209131cea835a3 |
| VAR | 1 | Not recorded | 68fa7a741e209131cea835a3 |
| Mo-VQGAN | 1.12 | Not recorded | 68fa6d741e209131cea3a058 |
| VILA-U | 1.25 | Not recorded | 68fa7a741e209131cea835a3 |
| ViT-VQGAN | 1.28 | Not recorded | 68fa6d741e209131cea3a058 |
| RQ-VAE | 1.3 | Not recorded | 68fa7a741e209131cea835a3 |
| TokenFlow | 1.37 | Not recorded | 68fa7a741e209131cea835a3 |
| Open-MAGVIT2-I-PT | 1.39 | Not recorded | 68fa76e01e209131cea6f87d |
| Open-MAGVIT2-I-PT | 1.67 | Not recorded | 68fa76e01e209131cea6f87d |
| VILA-U | 1.8 | Not recorded | 68fa7a741e209131cea835a3 |
| RQ-VAE | 1.83 | Not recorded | 68fa6d741e209131cea3a058 |
| Cosmos | 1.93 | Not recorded | 68fa76e01e209131cea6f87d |
| VAR | 2.09 | Not recorded | 68fa7a741e209131cea835a3 |
| LlamaGen | 2.19 | Not recorded | 68fa7a741e209131cea835a3 |
| LlamaGen | 2.47 | Not recorded | 68fa76e01e209131cea6f87d |
| Cosmos | 2.52 | Not recorded | 68fa76e01e209131cea6f87d |
| Open-MAGVIT2-I-PT | 2.55 | Not recorded | 68fa76e01e209131cea6f87d |
| RQ-VAE | 3.2 | Not recorded | 68fa7a741e209131cea835a3 |
| Show-o | 3.5 | Not recorded | 68fa76e01e209131cea6f87d |
| VQGAN | 3.64 | Not recorded | 68fa6d741e209131cea3a058 |
| Cosmos | 4.57 | Not recorded | 68fa76e01e209131cea6f87d |
| VQ-GAN | 4.98 | Not recorded | 68fa7a741e209131cea835a3 |
| VQGAN | 6.25 | Not recorded | 68fa6d741e209131cea3a058 |

### TokenFlow

rFID: 0.63. Result ID: 698b77f37a967eddb76ed9ac.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7a741e209131cea835a3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a741e209131cea835a3.md>)

[Implementation](<https://byteflow-ai.github.io/TokenFlow/>)

Recorded evaluation configuration: {"Resolution":"384x384","Downsample Ratio":14.2,"Residual Levels":15}. This record does not assert independent reproduction.

### Open-MAGVIT2-I-PT

rFID: 0.78. Result ID: 69944fb73a6c3148491c6e7d.

[Reporting paper](<https://www.sota2.com/research/paper/68fa76e01e209131cea6f87d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76e01e209131cea6f87d.md>)

[Implementation](<https://github.com/TencentARC/Open-MAGVIT2>)

Recorded evaluation configuration: {"Resolution":"Original Resolution","Quantizer Type":"LFQ","Training Data":"100M","Ratio":16,"Codebook Size":262144}. This record does not assert independent reproduction.

### LlamaGen

rFID: 0.94. Result ID: 698b77f37a967eddb76ed9a9.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7a741e209131cea835a3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa738a1e209131cea5d111.md>)

[Implementation](<https://github.com/FoundationVision/LlamaGen>)

Recorded evaluation configuration: {"Resolution":"384x384","Downsample Ratio":14.2,"Residual Levels":1}. This record does not assert independent reproduction.

### VAR

rFID: 1. Result ID: 698b77f37a967eddb76ed9a6.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7a741e209131cea835a3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73861e209131cea5cf19.md>)

[Implementation](<https://github.com/FoundationVision/VAR>)

Recorded evaluation configuration: {"Resolution":"256x256","Downsample Ratio":16,"Residual Levels":10}. This record does not assert independent reproduction.

### Mo-VQGAN

rFID: 1.12. Result ID: 698fa3c63a6c314849fabe25.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6d741e209131cea3a058.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d741e209131cea3a058.md>)

[Implementation](<https://github.com/CompVis/taming-transformers>)

Recorded evaluation configuration: {"Latent Size":"16x16x4","Num Z":"1024"}. This record does not assert independent reproduction.

### VILA-U

rFID: 1.25. Result ID: 698b77f37a967eddb76ed9aa.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7a741e209131cea835a3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77391e209131cea71cbd.md>)

Recorded evaluation configuration: {"Resolution":"384x384","Downsample Ratio":14.2,"Residual Levels":16}. This record does not assert independent reproduction.

### ViT-VQGAN

rFID: 1.28. Result ID: 698fa3c63a6c314849fabe23.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6d741e209131cea3a058.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6cba1e209131cea35a1f.md>)

Recorded evaluation configuration: {"Latent Size":"32x32","Num Z":"8192"}. This record does not assert independent reproduction.

### RQ-VAE

rFID: 1.3. Result ID: 698b77f37a967eddb76ed9a5.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7a741e209131cea835a3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6bf21e209131cea311a9.md>)

Recorded evaluation configuration: {"Resolution":"256x256","Downsample Ratio":16,"Residual Levels":4}. This record does not assert independent reproduction.

### TokenFlow

rFID: 1.37. Result ID: 698b77f37a967eddb76ed9a8.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7a741e209131cea835a3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a741e209131cea835a3.md>)

[Implementation](<https://byteflow-ai.github.io/TokenFlow/>)

Recorded evaluation configuration: {"Resolution":"256x256","Downsample Ratio":16,"Residual Levels":9}. This record does not assert independent reproduction.

### Open-MAGVIT2-I-PT

rFID: 1.39. Result ID: 69944fb73a6c3148491c6e7c.

[Reporting paper](<https://www.sota2.com/research/paper/68fa76e01e209131cea6f87d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76e01e209131cea6f87d.md>)

[Implementation](<https://github.com/TencentARC/Open-MAGVIT2>)

Recorded evaluation configuration: {"Resolution":"Original Resolution","Quantizer Type":"LFQ","Training Data":"100M","Ratio":16,"Codebook Size":16384}. This record does not assert independent reproduction.

### Open-MAGVIT2-I-PT

rFID: 1.67. Result ID: 69944fb73a6c3148491c6e79.

[Reporting paper](<https://www.sota2.com/research/paper/68fa76e01e209131cea6f87d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76e01e209131cea6f87d.md>)

[Implementation](<https://github.com/TencentARC/Open-MAGVIT2>)

Recorded evaluation configuration: {"Resolution":"Resize 256 x 256","Quantizer Type":"LFQ","Training Data":"100M","Ratio":16,"Codebook Size":262144}. This record does not assert independent reproduction.

### VILA-U

rFID: 1.8. Result ID: 698b77f37a967eddb76ed9a7.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7a741e209131cea835a3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77391e209131cea71cbd.md>)

Recorded evaluation configuration: {"Resolution":"256x256","Downsample Ratio":16,"Residual Levels":4}. This record does not assert independent reproduction.

### RQ-VAE

rFID: 1.83. Result ID: 698fa3c63a6c314849fabe24.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6d741e209131cea3a058.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6bf21e209131cea311a9.md>)

Recorded evaluation configuration: {"Latent Size":"8x8x16","Num Z":"16384"}. This record does not assert independent reproduction.

### Cosmos

rFID: 1.93. Result ID: 69944fb73a6c3148491c6e7a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa76e01e209131cea6f87d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa79ec1e209131cea7ffbc.md>)

Recorded evaluation configuration: {"Resolution":"Original Resolution","Quantizer Type":"FSQ","Ratio":16,"Codebook Size":64000}. This record does not assert independent reproduction.

### VAR

rFID: 2.09. Result ID: 698b77f37a967eddb76ed9ab.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7a741e209131cea835a3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73861e209131cea5cf19.md>)

[Implementation](<https://github.com/FoundationVision/VAR>)

Recorded evaluation configuration: {"Resolution":"384x384","Downsample Ratio":16,"Residual Levels":13}. This record does not assert independent reproduction.

### LlamaGen

rFID: 2.19. Result ID: 698b77f37a967eddb76ed9a3.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7a741e209131cea835a3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa738a1e209131cea5d111.md>)

[Implementation](<https://github.com/FoundationVision/LlamaGen>)

Recorded evaluation configuration: {"Resolution":"256x256","Downsample Ratio":16,"Residual Levels":1}. This record does not assert independent reproduction.

### LlamaGen

rFID: 2.47. Result ID: 69944fb73a6c3148491c6e75.

[Reporting paper](<https://www.sota2.com/research/paper/68fa76e01e209131cea6f87d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa738a1e209131cea5d111.md>)

[Implementation](<https://github.com/FoundationVision/LlamaGen>)

Recorded evaluation configuration: {"Resolution":"Resize 256 x 256","Quantizer Type":"VQ","Training Data":"70M","Ratio":16,"Codebook Size":16384,"Initial Training":"Pre-trained on ImageNet"}. This record does not assert independent reproduction.

### Cosmos

rFID: 2.52. Result ID: 69944fb73a6c3148491c6e7b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa76e01e209131cea6f87d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa79ec1e209131cea7ffbc.md>)

Recorded evaluation configuration: {"Resolution":"Original Resolution","Quantizer Type":"FSQ","Ratio":16,"Codebook Size":64000,"Source":"Reported by Cosmos"}. This record does not assert independent reproduction.

### Open-MAGVIT2-I-PT

rFID: 2.55. Result ID: 69944fb73a6c3148491c6e78.

[Reporting paper](<https://www.sota2.com/research/paper/68fa76e01e209131cea6f87d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76e01e209131cea6f87d.md>)

[Implementation](<https://github.com/TencentARC/Open-MAGVIT2>)

Recorded evaluation configuration: {"Resolution":"Resize 256 x 256","Quantizer Type":"LFQ","Training Data":"100M","Ratio":16,"Codebook Size":16384}. This record does not assert independent reproduction.

### RQ-VAE

rFID: 3.2. Result ID: 698b77f37a967eddb76ed9a4.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7a741e209131cea835a3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6bf21e209131cea311a9.md>)

Recorded evaluation configuration: {"Resolution":"256x256","Downsample Ratio":32,"Residual Levels":4}. This record does not assert independent reproduction.

### Show-o

rFID: 3.5. Result ID: 69944fb73a6c3148491c6e76.

[Reporting paper](<https://www.sota2.com/research/paper/68fa76e01e209131cea6f87d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b1f1e209131cea86ae5.md>)

[Implementation](<https://github.com/showlab/Show-o>)

Recorded evaluation configuration: {"Resolution":"Resize 256 x 256","Quantizer Type":"LFQ","Training Data":"35M","Ratio":16,"Codebook Size":8192}. This record does not assert independent reproduction.

### VQGAN

rFID: 3.64. Result ID: 698fa3c63a6c314849fabe22.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6d741e209131cea3a058.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6a371e209131cea26a4d.md>)

[Implementation](<https://git.io/JnyvK>)

Recorded evaluation configuration: {"Latent Size":"16x16","Num Z":"16384"}. This record does not assert independent reproduction.

### Cosmos

rFID: 4.57. Result ID: 69944fb73a6c3148491c6e77.

[Reporting paper](<https://www.sota2.com/research/paper/68fa76e01e209131cea6f87d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa79ec1e209131cea7ffbc.md>)

Recorded evaluation configuration: {"Resolution":"Resize 256 x 256","Quantizer Type":"FSQ","Ratio":16,"Codebook Size":64000}. This record does not assert independent reproduction.

### VQ-GAN

rFID: 4.98. Result ID: 698b77f37a967eddb76ed9a2.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7a741e209131cea835a3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6a371e209131cea26a4d.md>)

[Implementation](<https://git.io/JnyvK>)

Recorded evaluation configuration: {"Resolution":"256x256","Downsample Ratio":16,"Residual Levels":1}. This record does not assert independent reproduction.

### VQGAN

rFID: 6.25. Result ID: 698fa3c63a6c314849fabe21.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6d741e209131cea3a058.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6a371e209131cea26a4d.md>)

[Implementation](<https://git.io/JnyvK>)

Recorded evaluation configuration: {"Latent Size":"16x16","Num Z":"1024"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
