Evaluating AHIQ on traditional IQA datasets

IIGROUP / AHIQ

[CVPRW 2022] Attentions Help CNNs See Better: Attention-based Hybrid Image Quality Assessment Network

MIT License

72 stars 5 forks source link

Evaluating AHIQ on traditional IQA datasets #3

Open ch-andrei opened 2 years ago

ch-andrei commented 2 years ago

Question for the results in the original paper: How is the evaluation on traditional datasets (LIVE, CSIQ, TID) performed? Do you report average performance over K runs? The paper only mentions that datasets are split 60-20-20 train/val/test. Please add a more detailed description.

Buka-Xing commented 10 months ago

I have the same question, especially when I test the whole TID2013 dataset with the code 'test.py'. The results are totally unsatisfactory and the PLCC, and SRCC scores are very low. Can anyone teach me how to correctly test the results?