[CVPR2024] SeeSR: Towards Semantics-Aware Real-World Image Super-Resolution
Apache License 2.0
331
stars
14
forks
source link
According to the latest code provided, retrain, can't fully reproduce the metrics in the paper, please tell me the results in the paper, are there any settings that need extra attention? #41
DrealSR in the paper:
27.9
0.7628
0.3299
0.2363
151.88
6.4893
0.6005
65.11
0.6708
reproduce:
28.15
0.7832
0.3175
0.2332
164.45
6.867
0.4413
61.38
0.6167
It seems that the reproduced model has higher fidelity and much less generative power.
DrealSR in the paper: 27.9 0.7628 0.3299 0.2363 151.88 6.4893 0.6005 65.11 0.6708 reproduce: 28.15 0.7832 0.3175 0.2332 164.45 6.867 0.4413 61.38 0.6167 It seems that the reproduced model has higher fidelity and much less generative power.