- Rating
- 3(+1)
- Confidence
- 2(−1)
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
The paper presents Anc-Diff, a conditional diffusion framework for generating missing historical forms of Chinese characters from available forms in other scripts. It combines component-level pre-training on paired cross-script components, a separate condition encoder that injects features into the U-Net bottleneck, and a shared patch sampling strategy for component and character training. Evaluation uses Font-607, containing 607 characters across six scripts with training and test sets separated by character identity. Validation includes comparisons with eight baselines, image similarity metrics, expert ratings, ablations of pre-training, injection positions, and patch sampling, and paired significance tests for Seal Script generation. The reported improvements are strongest for Seal and Clerical scripts, while performance on earlier scripts is mixed.
Critical strengths
- Anc-Diff improves LPIPS and cosine similarity over all eight baselines for Seal and Clerical Script generation under the Regular Script input setting. Expert ratings also favor the method, and paired tests with Holm correction support the Seal Script improvements. These complementary results provide substantive evidence of better generation quality on the evaluated tasks.
- Pre-training on paired cross-script components offers a practical way to exploit reusable character structures when whole-character training data are limited. Table 3 shows improvements for both Seal and Clerical scripts, while the separation of character identities across training and testing, and of component-source characters from Font-607, strengthens the evidence for transfer to unseen characters.
Critical weaknesses
1.Experiments rely on a single dataset containing characters with available forms in all six scripts. It remains unclear whether the improvements extend to characters with incomplete historical records, limiting the evidence for the intended lost-character restoration setting. 2.The pretrained model uses additional component data and a different training schedule from the model trained from scratch. This makes it difficult to isolate the benefit of component-specific priors from additional data and optimization. A matched-budget comparison would strengthen the central contribution.
Questions for authors
Reference Critical Weaknesses.
5 comments
Author Response To Reviewer (2/2)
Q3 & W3. On source-script selection, early-script performance, and genuinely missing historical forms.
(1) Source-script selection.
For each target script, we evaluate generation from the other five source scripts, following the same protocol as Table 1. Table 7 reports each method's results using the source script with the lowest test LPIPS, with CosSim and SSIM evaluated under the same configuration. Source scripts are therefore selected based on test performance and may differ across methods.
(2) Early-script performance and comparison with CycleGAN.
In the Oracle Bone Inscription and Bronze Inscription generation tasks in Table 7, CycleGAN achieves lower LPIPS than our method. However, CycleGAN achieves these favorable scores because of its limited ability to transform the source images, as explained below.
CycleGAN and Anc-Diff use OBI and BI interchangeably as source and target. Direct comparison of the paired original OBI and BI images gives a CosSim of 0.7531 and an LPIPS of 0.3239, which are even better than the corresponding generated results. This indicates that favorable similarity scores can be obtained by leaving the source image unchanged, without performing cross-script translation. CycleGAN performs only limited transformations, which gives it an advantage.
As discussed in Section 4.4 and Figure 4, CycleGAN largely preserves Regular Script characteristics when generating Seal Script. Figure 12 in Appendix A.3 provides source–output–target comparisons for this task, further illustrating this pattern. The comparisons between Oracle Bone and Bronze scripts in Figures 13 and 14 in Appendix A.3 show similar source-preservation patterns.
Since these two scripts already share similar forms, limited changes to the source images may partly explain CycleGAN’s favorable LPIPS. On tasks with larger differences between source and target scripts, such as generating Seal Script from Regular Script, CycleGAN performs substantially worse than our method.
(3) Variation across training runs and statistical significance.
The source-preservation patterns discussed above help explain CycleGAN’s favorable LPIPS, but do not establish the statistical significance of the differences on Oracle Bone and Bronze scripts. We therefore do not claim statistically significant improvements on these two tasks.
For Seal and Clerical script generation, we report the mean and standard deviation over five independent training runs for Anc-Diff, BBDM, FontDiffuser, and OBSD. These results support the stability of our method and indicate that the performance gains are consistent across runs rather than specific to a particular training run.
| Target | Source | Method | CosSim ↑ | SSIM ↑ | LPIPS ↓ |
|---|---|---|---|---|---|
| SS | RS | BBDM | 0.7617±0.0056 | 0.5110±0.0123 | 0.2831±0.0049 |
| SS | RS | FontDiffuser | 0.7555±0.0067 | 0.4662±0.0095 | 0.3140±0.0069 |
| SS | RS | OBSD | 0.7697±0.0029 | 0.5031±0.0027 | 0.2814±0.0022 |
| SS | RS | Anc-Diff (Ours) | 0.8087±0.0011 | 0.5092±0.0019 | 0.2348±0.0005 |
| CS | RS | BBDM | 0.8016±0.0097 | 0.4650±0.0033 | 0.2922±0.0114 |
| CS | RS | FontDiffuser | 0.8155±0.0141 | 0.4501±0.0073 | 0.2758±0.0107 |
| CS | RS | OBSD | 0.8176±0.0027 | 0.4528±0.0018 | 0.2694±0.0013 |
| CS | RS | Anc-Diff (Ours) | 0.8511±0.0016 | 0.4590±0.0010 | 0.2394±0.0010 |
We also report paired significance tests using scores for all 180 individual test characters in Seal Script generation in Appendix A.5. Although we have not conducted significance tests for Oracle Bone and Bronze script generation, the repeated experiments and significance tests above provide additional evidence of the stability and effectiveness of our method.
(4) Evaluation of genuinely missing historical forms.
Regarding genuinely missing historical forms, we have explored such cases and discussed the generated forms with ancient Chinese character experts. However, without surviving historical evidence, these examples can be assessed for plausibility but cannot be verified as historically authentic forms. Our quantitative evaluation therefore uses characters with all six script forms available to provide ground truth for controlled evaluation.
We hope our responses and the revisions to the manuscript address all your concerns. If anything remains unclear or insufficiently addressed, please let us know so that we can respond further.
Author Response To Reviewer On Q3
We have previously explained why CycleGAN achieves relatively strong results on early scripts (Appendix A.3) and presented multi-run experimental results for some methods on the SS and CS generation tasks, demonstrating that our method's performance is not attributable to randomness. Here, we provide the mean and standard deviation over five independent training runs of our method across all tasks, as shown in the table below.
| Target | Source | Method | CosSim ↑ | SSIM ↑ | LPIPS ↓ |
|---|---|---|---|---|---|
| OBI | BI | Anc-Diff | 0.7236±0.0023 | 0.5119±0.0018 | 0.3489±0.0008 |
| BI | OBI | Anc-Diff | 0.7238±0.0029 | 0.4421±0.0023 | 0.3615±0.0028 |
| SS | WSS | Anc-Diff | 0.7935±0.0018 | 0.3584±0.0015 | 0.3320±0.0013 |
| SS | RS | Anc-Diff | 0.8087±0.0011 | 0.5092±0.0019 | 0.2348±0.0005 |
| CS | RS | Anc-Diff | 0.8511±0.0016 | 0.4590±0.0010 | 0.2394±0.0010 |
For each task, we tested five source scripts and reported the best result for each method. Even under this favorable setting, all baselines except CycleGAN perform worse than ours. Using the same source script would further degrade the baselines' performance. Therefore, under the same source inputs, our results remain statistically significantly superior to those of the baseline methods across both Oracle Bone and Bronze Script tasks. As shown in the table below.
| Method | Target: OBI (Source) | OBI CosSim↑ | OBI SSIM↑ | OBI LPIPS↓ | Target: BI (Source) | BI CosSim↑ | BI SSIM↑ | BI LPIPS↓ |
|---|---|---|---|---|---|---|---|---|
| CycleGAN | BI | 0.7217 | 0.4737 | 0.3377 | OBI | 0.7203 | 0.4403 | 0.3501 |
| WSS | 0.7029 | 0.4562 | 0.3728 | WSS | 0.7194 | 0.3977 | 0.3712 | |
| SS | 0.7093 | 0.4953 | 0.3431 | SS | 0.7218 | 0.4363 | 0.3538 | |
| CS | 0.6971 | 0.4397 | 0.4017 | CS | 0.7053 | 0.3620 | 0.3968 | |
| RS | 0.7107 | 0.4727 | 0.3620 | RS | 0.7189 | 0.3948 | 0.3676 | |
| Pix2pix | BI | 0.6580 | 0.5489 | 0.3551 | OBI | 0.6498 | 0.4754 | 0.3897 |
| WSS | 0.6720 | 0.5450 | 0.3551 | WSS | 0.6647 | 0.4799 | 0.3822 | |
| SS | 0.6510 | 0.5465 | 0.3606 | SS | 0.6688 | 0.4812 | 0.3820 | |
| CS | 0.6528 | 0.5389 | 0.3605 | CS | 0.6598 | 0.4785 | 0.3924 | |
| RS | 0.6818 | 0.5439 | 0.3510 | RS | 0.6888 | 0.4892 | 0.3775 | |
| DRIT++ | BI | 0.7029 | 0.4520 | 0.3545 | OBI | 0.6678 | 0.5010 | 0.4234 |
| WSS | 0.6558 | 0.5677 | 0.4294 | WSS | 0.7036 | 0.4619 | 0.3820 | |
| SS | 0.6919 | 0.4731 | 0.3586 | SS | 0.6992 | 0.4415 | 0.3717 | |
| CS | 0.6839 | 0.4761 | 0.4082 | CS | 0.6896 | 0.4091 | 0.4130 | |
| RS | 0.7102 | 0.4693 | 0.3588 | RS | 0.6850 | 0.4709 | 0.3985 | |
| BBDM | BI | 0.7011 | 0.5334 | 0.3606 | OBI | 0.6963 | 0.4786 | 0.3839 |
| WSS | 0.6428 | 0.5558 | 0.4117 | WSS | 0.7036 | 0.4749 | 0.3765 | |
| SS | 0.6942 | 0.5183 | 0.3677 | SS | 0.6005 | 0.5164 | 0.4552 | |
| CS | 0.6864 | 0.5164 | 0.3966 | CS | 0.6760 | 0.4890 | 0.4079 | |
| RS | 0.6629 | 0.5305 | 0.4087 | RS | 0.6720 | 0.4933 | 0.3856 | |
| FontDiffuser | BI | 0.7144 | 0.4701 | 0.3656 | OBI | 0.6803 | 0.5114 | 0.4017 |
| WSS | 0.6935 | 0.4406 | 0.3823 | WSS | 0.6840 | 0.4831 | 0.3869 | |
| SS | 0.7103 | 0.4416 | 0.3757 | SS | 0.7026 | 0.4876 | 0.3932 | |
| CS | 0.6974 | 0.4567 | 0.3870 | CS | 0.6795 | 0.4864 | 0.4063 | |
| RS | 0.6981 | 0.4391 | 0.3804 | RS | 0.6927 | 0.4754 | 0.3871 | |
| OBSD | BI | 0.6907 | 0.5348 | 0.3846 | OBI | 0.6954 | 0.4674 | 0.3940 |
| WSS | 0.6582 | 0.5468 | 0.4223 | WSS | 0.6697 | 0.4777 | 0.4199 | |
| SS | 0.6776 | 0.5450 | 0.3903 | SS | 0.6850 | 0.4841 | 0.3937 | |
| CS | 0.6715 | 0.5364 | 0.4091 | CS | 0.6889 | 0.4636 | 0.4005 | |
| RS | 0.6896 | 0.5269 | 0.3870 | RS | 0.7026 | 0.4525 | 0.3948 | |
| Anc-Diff | BI | 0.7223 | 0.5107 | 0.3486 | OBI | 0.7259 | 0.4402 | 0.3647 |
| WSS | 0.7073 | 0.5139 | 0.3626 | WSS | 0.7165 | 0.4441 | 0.3784 | |
| SS | 0.7210 | 0.5013 | 0.3536 | SS | 0.7213 | 0.4426 | 0.3620 | |
| CS | 0.7049 | 0.5018 | 0.3664 | CS | 0.7165 | 0.4321 | 0.3787 | |
| RS | 0.7142 | 0.5008 | 0.3607 | RS | 0.7241 | 0.4196 | 0.3805 |
Author Response To Reviewer (1/2)
Thank you for the detailed and constructive feedback. Your comments help us improve the experimental comparisons and explain our methodological contributions and limitations more clearly. We have revised the manuscript accordingly and respond to your concerns below.
Q1 & W1. On component-specific gains versus generic cross-script pre-training.
In the original Table 3, the full model uses 300 pre-training and 100 character-training epochs, and the variant trains on characters for 300 epochs. These schedules are selected based on the experimental performance of each variant. This ablation demonstrates the benefit of adding component pre-training, but does not isolate the benefit of component-specific knowledge.
To distinguish component-specific gains from the benefit of additional pre-training, we add two variants.
The first removes component pre-training and retains the same 100 epochs of character training as the full model (w/o Pre. (100 epochs)).
The second replaces paired components with paired characters for generic cross-script pre-training, followed by the same 100 epochs of character training (Character pre-training). Since the component and character training sets contain 337 and 427 paired samples, respectively, we use 237 epochs of character pre-training based on 300 × 337 / 427 ≈ 237. With the same number of patches sampled per image, this setting approximately matches the full model in both patch counts and optimizer updates.
We also retain the original variant without component pre-training, which uses 300 epochs of character training (w/o Pre. (300 epochs)).
The experimental results are presented below and have been added to Section 4.6.1 and Table 3 of the revised manuscript.
The comparison with w/o Pre. (100 epochs) demonstrates the benefit of adding component pre-training. The comparison with Character pre-training under approximately matched optimizer updates and patch counts further shows that pre-training with paired cross-script components is more effective than generic cross-script pre-training with paired characters.
Q2 & W2. On comparisons with OBSD and the injection-position ablation.
(1) Comparisons with OBSD
We retain OBSD’s original sampling procedure. In the original pipeline, the second-stage refinement converts the font style of the first-stage output to the target style. In our implementation, the diffusion stage is trained directly to generate the target style. We therefore incorporate target-style generation into the diffusion stage and do not apply a separate FontDiffuser refinement stage.
We use the same backbone architecture and character training data as OBSD, with matched training updates to ensure a fair comparison. Our method further incorporates component priors, a sampling strategy, and a decoupled conditioning design. These task-specific designs improve generation quality. The additional component data used for pre-training are explicitly reported. The Character pre-training variant added in Q1 also outperforms OBSD without using additional component data.
(2) injection-position ablation
All variants in Table 4 use the same independent condition encoder, consisting of stacked convolutional layers. At each injection position, the conditional feature map at the corresponding spatial scale is directly concatenated with the target feature map. Concatenation itself introduces no learnable parameters, and no additional projection or attention module is added to favor the bottleneck variant.
The comparison therefore retains the same encoder architecture and fusion operation while changing the injection position. However, feature dimensions differ across U-Net scales, so the convolutional layers following concatenation may have different parameter counts. We have added these implementation details in the revised description of Section 4.6.2.