shrubb

nohup tensorboard --logdir . --samples_per_plugin scalars=500,images=90 --port 8091 --host 0.0.0.0 > tensorboard-log.txt 2>&1 &

shrubb

import random

def partition(array, start, end, pivot_idx):
    pivot = array[pivot_idx]
    array[pivot_idx], array[end - 1] = array[end - 1], array[pivot_idx]

    left, right = start, end - 2
    # invariant:
    # array[< left] < pivot
    # array[> right] >= pivot

shrubb

#include <cstdio>
#include <cstdlib>
#include <iostream>
#include <algorithm>
#include <vector>
#include <map>
#include <set>
#include <queue>
#include <cstring>
#include <cmath>

shrubb

# usage:

# $ ls
# 2.23.1.jpeg   2.23.jpg      DSC_0383.JPG  DSC_0411.JPG  DSC_0439.JPG  DSC_0467.JPG
# DSC_0496.JPG  DSC_0524.JPG  DSC_0553.JPG  DSC_0582.JPG  DSC_0610.JPG  DSC_0638.JPG
# DSC_0384.JPG  DSC_0412.JPG  DSC_0440.JPG  DSC_0468.JPG  DSC_0497.JPG  DSC_0525.JPG
# ... ... ...
# DSC_0611.JPG  DSC_0639.JPG  DSC_0357.NEF  DSC_0385.JPG  DSC_0413.JPG  DSC_0441.JPG
# DSC_0554.JPG  DSC_0583.JPG  DSC_0358.NEF  DSC_0386.JPG  DSC_0414.JPG  DSC_0442.JPG
# $ ./../rename.sh

shrubb

LIBS_ROOT=/nfs/hpc-4/hdd-raid/e_burkov/Libraries

# CMake variables (CMake will detect them)
CUDNN_ROOT=$LIBS_ROOT/cudnn-11.3
export CUDNN_LIBRARY_PATH=$CUDNN_ROOT/lib64/libcudnn.so
export CUDNN_LIBRARY=$CUDNN_ROOT/lib64
export CUDNN_INCLUDE_PATH=$CUDNN_ROOT/include
export CUDNN_INCLUDE_DIR=$CUDNN_ROOT/include

# Environment variables that will be detected too but aren't CMake variables

shrubb

require 'cunn'

torch.setdefaulttensortype('torch.FloatTensor')

torch.manualSeed(666)
cutorch.manualSeed(666)

-- change these as needed
-- no automated testing, sorry :(
local nInputPlane, nOutputPlane, inputHeight, inputWidth = 128, 1, 52, 52

shrubb

def get_next_token_probabilities(curr_token_sequence, z, embed_fn, decode_fn):
    curr_tokens_embeddings = embed_fn([curr_token_sequence])
    return decode_fn(curr_tokens_embeddings, z)[0][-1]

def beam_search_step(current_token_sequence, get_next_token_probabilities, depth=3, width=4):
    """
    returns: next token guess
    """
    candidates = [(1.0, [])] # (probability, candidate_appendix)
    

shrubb

# encoding: utf-8
import sys
import csv
import re

def preprocess_text(text):
    text = text.replace('\t', '')
    text = text.replace('ё' , 'е')
    text = text.replace('…' , '...')
    for c in '–—–':

shrubb

shrubb

Interactive Face Recognition with GPU

To provide an example of how OpenCV's CUDA module can be used in Torch, we have implemented interactive face recognition in this application.

Important warning

In this very example, the recognition quality may be rather poor as, in fact, the OpenFace's face recognition framework is intended to work in a more complicated way. Here, we didn't locate facial landmarks and estimate head pose, although this is an essential part of the pipeline. Keep this in mind when working with this task, and consult OpenFace's website for further information.

How to use it