李鼎权, Dingquan Li

@Rubiel1 Your implementation of the second way is wrong. You should be aware of the block-based calculation for SSIM map. It is not `(img1-mu1).pow(2)`, actually, you need to calculate E[(img1_block_ij-mu1_ij).^2]...

@williamFalcon Thanks. I may have a try and update the repo when I am free.

@lucky1214 Actually, it is 29.08dB in Table 2 of Nah's paper.

Another machine (that reproduced the randomness) info ```text ==========System========== Linux-4.4.0-176-generic-x86_64-with-debian-buster-sid DISTRIB_ID=Ubuntu DISTRIB_RELEASE=18.04 DISTRIB_CODENAME=bionic DISTRIB_DESCRIPTION="Ubuntu 18.04.4 LTS" 3.7.10 | packaged by conda-forge | (default, Feb 19 2021, 16:07:37) [GCC 9.3.0] ==========Pytorch==========...

I found [src/3rdparty/cudf/detail/utilities/device_atomics.cuh](https://github.com/NVIDIA/MinkowskiEngine/blob/b71caef9b45ab32c02d2c1b5f48f13c1eb5df526/src/3rdparty/cudf/detail/utilities/device_atomics.cuh) and `atomicAdd`. Is this randomness introduced by the routine **cublasSetAtomicsMode()**？Can I easily choose another rountine for a deterministic behavior on GPU training? Thanks! cuBLAS doc: [For some...

See README.md https://pan.baidu.com/s/1LS5GUmknqB_zUyEJTcUCGg?pwd=cu9j

Wow~Nice work on the VQA benchmark dataset! Congratulations for the acceptance of this work by NeurIPS Datasets and Benchmarks Track. Thank you for the effort and thanks for reporting the...

> CC @bradh > > @lidq92 HEIF creation capabilities have just landed in GDAL master (3.11.0dev) a few days ago. Did you try to build it from source? My current...

@bradh I appreciate the update and would be happy to assist with relevant work. However, I want to mention that my skills in C/C++ programming are somewhat limited. I will...

> [@lidq92](https://github.com/lidq92) I have also been studying remote sensing image compression (16-bit depth, 4 bands) recently, and I would like to know how to use VTM to process this type...