EO-NeRF data

[jieeeeeeeeeee]

EO-NeRF is an exciting project. The EO-NeRF paper mentions, 'The input data and EO-NeRF results are released to contribute to the creation of a NeRF benchmark for multi-date satellite images.' Could you confirm if the data used in EO-NeRF will be open-sourced? Currently, the [Data] link on the EO-NeRF project website is inaccessible.

[rogermm14]

Hello @jieeeeeeeeeee , thank you for the reminder :) The [Data] link is now updated and you can access the IARPA areas of interest discussed in EO-NeRF. The rest of areas are the same as in SatNeRF.

[jieeeeeeeeeee]

Hello @rogermm14, thank you for your response! The data from the EO-NeRF project is available for download. I appreciate your assistance in resolving this matter :)

[jieeeeeeeeeee]

Another question, IARPA Data includes 49 WorldView-3 images. In our project, we have found that the 3D reconstruction results using all these images are not ideal. Our speculation suggests two possible reasons:

1. Some images among the 49 have suboptimal quality, such as excessively skewed perspectives or dim radiometric conditions, leading to a decrease in pose accuracy during global Bundle Adjustment (BA).
2. Additional preprocessing may be necessary, similar to the approach used in the DFC2019 dataset, involving atmospheric correction or Radiometric Normalization to correct radiometric information.

Do you have any suggestions, for example, on how to filter the images? any advice you could give would be much appreciated. :raised_hands:

[fanzz1208]

您好，感谢您的提醒 :) [数据] 链接现已更新，您可以访问 EO-NeRF 中讨论的 IARPA 感兴趣领域。其余区域与 SatNeRF 中的区域相同。

We did not find “IARPA_001_DSM.txt” “IARPA_002_DSM.txt” “IARPA_003_DSM.txt”in the folder "SatNeRF_IARPA/Truth", just like DFC2019

How to determine AOI？

[rogermm14]

Hi @fanzz1208,

Thank you for the observation. You can find the AOI of the IARPA DSMs in the metadata of the .tif files. E.g. gdalinfo IARPA_001_DSM.tif in the command line or using the rasterio library in Python.

As a workaround, you can use this code to load the IARPA AOI information in the same format as in JAX:

    if "JAX" in aoi_id:
        # For JAX we need to load the AOI from the .txt files 
        gt_roi_path = os.path.join(gt_dir, "{}_DSM.txt".format(aoi_id))
        assert os.path.exists(gt_roi_path), f"{gt_roi_path} not found"
        gt_roi_metadata = np.loadtxt(gt_roi_path)
    else:
        # For IARPA we need to load the AOI from the .tif metadata 
        src = rasterio.open(gt_dsm_path)
        gt_roi_metadata = np.array([src.bounds.left, src.bounds.bottom, min(src.height, src.width), src.res[0]])
        del src