search

UARK-AICV / AerialFormer

[Remote Sensing] AerialFormer: Multi-resolution Transformer for Aerial Image Segmentation
https://www.mdpi.com/2072-4292/16/16/2930
39 stars 4 forks source link

Val/Test subset #2

Open clementveyer opened 3 months ago

clementveyer commented 3 months ago

Hi! In your preprocessing code, you create a “val” folder, but as I understand it, it corresponds to the usual “test” subset used for benchmark (2_13, 2_14, 3_13, …).

Then with what data do you evaluate your model during training ? Do you split your “train” subset into validation and train data?

Thanks for the work !

advaghasiya commented 3 months ago

Subject: GLIBC version compatibility issue with AerialFormer Singularity container Dear AerialFormer maintainers, I'm encountering a GLIBC version compatibility issue when trying to run the AerialFormer training script using the provided Singularity container. Here are the details:

Issue: The Singularity container requires GLIBC 2.34, but my host system has GLIBC 2.35. Error message: CopyImportError: /lib/x86_64-linux-gnu/libc.so.6: version `GLIBC_2.34' not found (required by /.singularity.d/libs/libGLdispatch.so.0)

Host system details:

OS: Ubuntu (version: [your Ubuntu version]) GLIBC version: 2.35 (Ubuntu GLIBC 2.35-0ubuntu3.8)

Reproduction steps:

e.g., bash tools/singularity_train.sh configs/aerialformer/aerialformer_small_512x512_6_potsdam.py

Expected behaviour: The training script should run without GLIBC version conflicts.

Could you please provide guidance on how to resolve this issue? Possible solutions might include: a) Providing a Singularity container compatible with a wider range of GLIBC versions, including 2.35. b) Updating the container's base image to one that's compatible with GLIBC 2.35. c) Providing instructions on how to modify the Singularity definition file to use a compatible base image. Any assistance or workarounds you can offer would be greatly appreciated. Thank you for your time and support. Best regards,