AruniRC

This walkthrough describes setting up Detectron (3rd party pytorch implementation) and Graph Conv Net (GCN) repos on the UMass cluster Gypsum. Most commands are specific to that setting.

Gypsum environment

$ module list
Currently Loaded Modulefiles:
  1) slurm/16.05.8                         3) hdf5/1.6.10                           5) gcc5/5.4.0                            7) cudnn/5.1
 2) openmpi/gcc/64/1.10.1 4) fftw2/openmpi/open64/64/float/2.1.5 6) cuda80/toolkit/8.0.61 8) hdf5_18/1.8.17

AruniRC

import numpy as np
import itertools
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.nn.init as init
from torch.autograd import Variable
from torch.autograd import Function
from scipy.spatial.distance import pdist

AruniRC

    def forward(self, x):
        x = self.features(x)
        [bs, ch, h, w] = x.shape
        x = x.view(bs, ch, -1).transpose(2, 1)
        # x.register_hook(self.save_grad('x'))

        # Gram Matrix NxN for the N input features "x"
        K = x.bmm(x.transpose(2, 1))
        K = x * x;   # < --- IS THIS CORRECT for 1st order features????

AruniRC

Intro Vim

• Intro and setup for IDE users: https://blog.stevenocchipinti.com/2014/03/17/introduction-to-vim-for-ide-users/
• Quick list: https://github.com/hackjutsu/vim-cheatsheet
• Quick reference card for print-out: http://tnerual.eriogerg.free.fr/vimqrc.pdf

Vim for Python

• Tabs and Spaces: https://linuxhint.com/vim-python-development/

AruniRC

	  ---			-----------------			----
	  Map			Quick Explanation			Link
	  ---			-----------------			----
<	 <F1>	Causes Netrw to issue help
	 <cr>	Netrw will enter the directory or read the file      |netrw-cr|
	 <del>	Netrw will attempt to remove the file/directory      |netrw-del|
	 <c-h>	Edit file hiding list                                |netrw-ctrl-h|
	 <c-l>	Causes Netrw to refresh the directory listing        |netrw-ctrl-l|
	 <c-r>	Browse using a gvim server                           |netrw-ctrl-r|
	 <c-tab> Shrink/expand a netrw/explore window                |netrw-c-tab|

AruniRC

if self.distill:
  soft_target = Variable(data[2].cuda())
  distill_loss = torch.mean(torch.sum(- nn.Softmax(dim=1)(soft_target/self.T) * nn.LogSoftmax(dim=1)(out_data/self.T), 1))  
  loss += self.lbda*distill_loss
  self.writer.add_scalar('train/distill_loss', distill_loss, i_acc+i+1)

AruniRC

General walkthrough for caffe on CentOS: http://www.aymeric.net/journal/2015/07/tutorial-install-caffe-on-centos-7/

Makefile

Make sure that the Python include directories actually exist (these differ subtly from Ubuntu to CentOS).

CUSTOM_CXX := g++

AruniRC

## Refer to http://caffe.berkeleyvision.org/installation.html
# Contributions simplifying and improving our build system are welcome!

# cuDNN acceleration switch (uncomment to build with cuDNN).
USE_CUDNN := 1

# CPU-only switch (uncomment to build without GPU support).
# CPU_ONLY := 1

# uncomment to disable IO dependencies and corresponding data layers

AruniRC

Getting docker running on CentOS server

   sudo systemctl start docker
   sudo systemctl status docker
   sudo systemctl enable docker
   sudo usermod -aG docker $(whoami)

AruniRC

Multi-Class Detection with RetinaNet

Location

On gypsum:

/mnt/nfs/scratch1/arunirc/data/MS-COCO_cls-1_hard-neg