- Rating
- 2
- Confidence
- 2
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper addresses post-hoc correction of degraded outputs from autoregressive image generators and proposes Acceptance-Gated Refinement (AGR), a two-level framework combining an image-level router with a token-level acceptance gate. The router uses refiner-derived scores to decide whether to preserve an input unchanged or refine it, while the gate selectively accepts proposed token replacements based on their score gains. Experiments on FFHQ, CelebA-HQ, and ImageNet demonstrate improvements in the quality of degraded samples while largely preserving well-tuned outputs.
Critical strengths
1.The proposed method is simple, intuitive, and easy to understand.
- The paper provides reasonably comprehensive internal validation on FFHQ, CelebA-HQ, and ImageNet, covering well-tuned samples, degraded samples, and unlabeled mixtures.
Critical weaknesses
-
The methodological novelty appears limited. Iterative refinement, confidence-based token selection, and quality-guided acceptance have been explored in prior work. The paper does not sufficiently clarify its technical advance beyond combining these established ideas into a two-level decision framework.
-
The application scope is narrow, with experiments mainly focusing on correcting degraded autoregressive image samples. Many degradations are artificially introduced, and their prevalence in practical generation remains unclear. Comparisons with independent external baselines are also limited. I strongly recommend that the authors include more baselines.
Questions for authors
See the weaknesses.
3 comments
Also here are more baseline
We added five, one per row of Table G2:
- the generator's own likelihood as router, deployed and full precision (Table R1-3);
- compute-matched best-of-N (Reiviewer-2);
- a fixed τ chosen per condition (G1);
- one fixed τ for every image on GPT-XL (G1b);
- SDEdit with a pretrained DiT-XL/2-256 (R3-2).
They come on top of the rate-matched random gate and the ungated limit in the paper, and the Token-Critic-style confidence rule (R2-1b).
Table R3-2. SDEdit with a pretrained DiT-XL/2-256 vs AGR (ImageNet GPT-B, first 5k test images; s = SDEdit strength; "kept" = mean Inception-feature cosine similarity between input and output; two unrelated same-class images score 0.72).
| no repair | AGR | SDEdit s=0.2 | s=0.4 | s=0.6 | s=0.4 routed by AGR | DiT sampled directly | |
|---|---|---|---|---|---|---|---|
| FID tuned (cfg 2) | 11.10 | 11.09 | 10.68 | 9.95 | 9.68 | 11.09 | 10.02 |
| FID mild (cfg 1, t=1.2) | 55.99 | 47.92 | 42.98 | 18.68 | 9.04 | 32.11 | 10.02 |
| FID severe (cfg 1, t=1.5) | 98.60 | 61.51 | 79.35 | 27.22 | 10.28 | 27.46 | 10.02 |
| seconds per edited image | – | 0.114 | 0.114 | 0.205 | 0.296 | 0.205 | 0.686 per sample |
| kept, tuned | 1.000 | 0.999 | 0.892 | 0.804 | 0.745 | 0.999 | – |
| kept, mild | 1.000 | 0.873 | 0.796 | 0.667 | 0.606 | 0.823 | – |
Table G1b. GPT-XL mixtures: AGR vs one fixed τ for every image (label-free baseline; 10k per mixture; per-condition columns are the 5k halves used, biased upward vs 10k).
| policy | tuned + severe | tuned + mild | tuned (5k) | mild (5k) | severe (5k) |
|---|---|---|---|---|---|
| no repair | 51.64 | 20.71 | 9.81 | 61.64 | 143.40 |
| one τ=3 for all | 40.74 | 17.87 | 9.84 | 53.62 | 116.59 |
| one τ=2 for all | 29.07 | 16.72 | 10.33 | 48.34 | 86.28 |
| AGR | 26.69 | 17.59 | 9.77 | 54.39 | 83.09 |
| AGR, τ=2 open branch | 28.26 | 16.89 | 9.81 | 51.64 | 86.33 |
The authors have addressed some of my concerns, but I still have reservations about the paper’s methodological novelty and scope of applicability. I therefore maintain my original rating.
We thank the reviewer for the careful reading, and here are some experiments we results we add for rebuttal.
R3 — W1. Limited methodological novelty
Here is the novelty comparison table, same with R2-1, so based on the table. The contribution is knowing which images to skip, without any label, from 25 samples, and having that judgment carry over to damage and generators it never saw.
Table R3-1. What is new relative to prior work.
R3 — W2. Narrow scope; artificial degradations; prevalence
The paper already repairs five frozen backbones from 111M to 1.4B parameters: LlamaGen GPT-B, GPT-L and GPT-XL (384²), a 2021 VQGAN transformer, and the random-order RandAR-L. They cover two tokenizer families, 256² and 384² resolution, raster and random order, and three datasets (FFHQ, CelebA-HQ, ImageNet). The rebuttal adds GPT-XXL (Table G1), a real quantizer (below), and the router wrapping a non-AR repairer (SDEdit, Table R3-2).
And we think the degradations are practical. GPTQ quantization is how people serve large models cheaply. Speculative decoding is how they speed up AR generation. Temperature and guidance are user-facing settings. These aren’t artificial.
Also the router scales with the damage. Fit only on temperature, it opens proportionally on real GPTQ quantization: a lot when GPTQ hurts, barely at all when GPTQ is harmless. Repairing everything when it’s harmless would cost 8–11 FID. The router avoids that. The router also abstains where repair would not help (no CFG, truncation).
Table G1 Full method on stronger generators
Table R3-1. Deployment settings (FID, lower is better).