Siddhartha Gairola comments

Results 9 comments of


                                            Siddhartha Gairola

Training is very slow

Thank you for your reply. I am currently using a Batch-Size of 20 with 4 GPUs and training the model end to end. I have 500K pairs for training. It...

Adding functionality to compute principal curvatures.

@jacopoantonello let me know if this seems good? Thanks.

Derivatives?

@realWDC could you please share the above. @jacopoantonello @realWDC also is there an existing implementation to compute the Curvature at a given point? Thanks!

multi lang, each cc into a new file

@cfsmp3 I would like to work on this.

In file fit_rotations_mex.cpp #endif should be outside parenthesis

Mr. Jacobson could this be clarified? After making that change it works locally on my machine.

Running the mex files in Scilab

Hi! Yes, Scilab is an open-source software similar to MATLAB and OCTAVE. It has it's own version of the MEX_API. I needed some help with the compilation of this toolbox...

Running the mex files in Scilab

Thanks a lot! Yeah will do so. Cheers!

Requirements.txt has modules with missing versions

hi @mbanani, Actually later on only the preprocessing code works (which I could send a pull request for). However, the training / eval still run into a lot of issues...

[MMdetection 3] How should I use gradient accumulation correctly

For your question - Ir is 0.01 when batch-size is 8*2=16. Now when running on single-gpu with batch-size=8 and accumulative_counts=2, then you will use the same lr = 0.01 as...