search

hexiao0275 / S2ADet

[TGRS 2023] The official repo for the paper "Object Detection in Hyperspectral Image via Unified Spectral-Spatial Feature Aggregation".
GNU Affero General Public License v3.0
25 stars 6 forks source link

HOD-1 dataset problem #7

Closed Bangyan-Hu closed 4 months ago

Bangyan-Hu commented 6 months ago

Hello, when I tried to reproduce the results of the article, I obtained relatively similar results on the HOD3K data set, but the effect on the HOD-1 data set was very poor, and it was almost impossible to predict any effective targets. So I have two questions: First of all, since you only released the original multispectral data of the HOD-1 data set and did not disclose the processed SA and SE data, can you please release the processed data set? Second, do you still use the same parameters hyp.finetune.yaml on the HOD-1 data set?

hexiao0275 commented 6 months ago

Hi. 1. I have uploaded the link to the HOD-1 processed dataset, which can be seen in Readme.md. Also, the generated code can be found in sa_se_image_generate.zip.

  1. Yes, HOD-1 is also using the same setup in order to ensure the fairness of the experiments