RepTok Prior vs GW Direct BGE On COCO
GW direct BGE
37.8M
- LPIPS ↓
- 0.7015
- CLIP ↑
- 0.2864
- FID ↓
- 77.60
GW short_T5 grounded
38.3M
- LPIPS ↓
- 0.6959
- CLIP ↑
- 0.2888
- FID ↓
- 67.09
GW 12h direct BGE
51.0M
- LPIPS ↓
- 0.6972
- CLIP ↑
- 0.2864
- FID ↓
- 75.61
GW 12h activated short_T5
51.5M
- LPIPS ↓
- 0.6953
- CLIP ↑
- 0.2916
- FID ↓
- 68.93
RepTok prior 796M
796.4M
- LPIPS ↓
- 0.7215
- CLIP ↑
- 0.3034
- FID ↓
- 65.89
RepTok prior 37M
36.8M
- LPIPS ↓
- 0.7332
- CLIP ↑
- 0.2774
- FID ↓
- 81.88
RepTok prior BGE memory
850.1M
- LPIPS ↓
- 0.7245
- CLIP ↑
- 0.2903
- FID ↓
- 70.26
Metric Barplots
Parameter Bubble Plot
Side-by-Side Generations
12h Activated-Start GW Samples
RepTok Prior Breakdown
| image/time/text projections | 14.8M |
| latent transformer blocks | 780.6M |
| learned helper/null/position tokens | 0.1M |
| output head | 1.0M |
| RepTok prior 37M: image/time/text projections | 2.2M |
| RepTok prior 37M: latent transformer blocks | 34.2M |
| RepTok prior 37M: learned helper/null/position tokens | 0.0M |
| RepTok prior 37M: output head | 0.3M |
| RepTok prior BGE memory: image/time/text projections | 14.1M |
| RepTok prior BGE memory: latent transformer blocks | 780.6M |
| RepTok prior BGE memory: learned helper/null/position tokens | 0.0M |
| RepTok prior BGE memory: output head | 1.0M |