Remote Sensing Image Captioning on Geo170K
0.6098BLEU-1FFT
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| FFTPost-training Adaptation=Full Fine-Tuning2026.04 | 0.6098 | 0.5526 | 2.3109 | 0.5627 | 0.4667 | 2,172.37 | 0.5693 | 0.6451 | |
| S-GRPO(wo CGI)Post-training Adaptation=Ablated Unified Post-Training without CGI2026.04 | 0.6086 | 0.5525 | 2.3087 | 0.5627 | 0.5111 | 2,222.08 | 0.6231 | 0.7098 | |
| S-GRPOPost-training Adaptation=Unified Post-Training2026.04 | 0.6075 | 0.5525 | 2.3008 | 0.5626 | 0.5211 | 2,322.45 | 0.643 | 0.7285 | |
| SFT → GRPONPost-training Adaptation=Supervised Fine-Tuning then GRPON2026.04 | 0.6012 | 0.5504 | 2.2896 | 0.5629 | 0.4767 | 2,199.34 | 0.5801 | 0.6594 | |
| RCPAPost-training Adaptation=Reference-based Constraint Policy Alignment2026.04 | 0.5998 | 0.5501 | 2.2821 | 0.5623 | 0.5122 | 2,315.37 | 0.6414 | 0.7278 | |
| CFFTPost-training Adaptation=Catastrophic Forgetting Free Fine-Tuning2026.04 | 0.5719 | 0.5091 | 1.8827 | 0.5079 | 0.5078 | 2,199.59 | 0.5888 | 0.6676 | |
| GCPOPost-training Adaptation=Gradient-based Constraint Policy Optimization2026.04 | 0.5689 | 0.5306 | 2.0324 | 0.5487 | 0.5089 | 2,203.65 | 0.5996 | 0.6857 | |
| GRPONPost-training Adaptation=GRPO with normalization2026.04 | 0.4086 | 0.3431 | 0.3543 | 0.335 | 0.5122 | 2,320.54 | 0.6414 | 0.7062 | |
| BASEPost-training Adaptation=Pre-trained Base Model2026.04 | 0.3859 | 0.3014 | 0.274 | 0.2901 | 0.5122 | 2,333.36 | 0.6211 | 0.7038 | |
| DAPOPost-training Adaptation=Direct Alignment Policy Optimization2026.04 | 0.3835 | 0.3189 | 0.2776 | 0.2989 | 0.5111 | 2,319.25 | 0.6285 | 0.711 | |
| GRPOPost-training Adaptation=Group Relative Policy Optimization2026.04 | 0.3799 | 0.3113 | 0.2661 | 0.2878 | 0.5022 | 2,346.08 | 0.6373 | 0.7131 | |
| PEFTPost-training Adaptation=Parameter-Efficient Fine-Tuning2026.04 | 0.0901 | 0.1192 | 0.0009 | — | 0.6089 | 2,449.59 | 0.536 | 0.6535 |