Tom Runia tomrunia

Caffe Classification

Classification Speed Benchmark

Table lists average classification times per image (milliseconds). Averages are computed over classification of 100 images from MS-COCO validation set. For the VGG networks a very high variance in classification times was observed, some images were classified fast while most of them took more processing time that other network configurations.

AlexNet

Classification done in 79.1ms [avg = 79.1ms] Probabilities and labels: n02124075 Egyptian cat (0.31) n02123159 tiger cat (0.18) n02123045 tabby, tabby cat (0.12) n02119022 red fox, Vulpes vulpes (0.11) n02085620 Chihuahua (0.04)

	{
	"icfdetector" :
	{
	"boost_type" : "WaldBoost",
	"detection_max_size" : 200,
	"detection_min_size" : 30,
	"feature_candidates" : 40000,
	"weak_count" : 600,
	"min_feature_size" : 25,
	"model_height" : 64,

	import numpy as np
	from tensorflow.python.summary.event_accumulator import EventAccumulator

	import matplotlib as mpl
	import matplotlib.pyplot as plt

	def plot_tensorflow_log(path):

	# Loading too much data is slow...
	tf_size_guidance = {

	#!/usr/bin/python
	import os

	# YouTube video searching API
	from apiclient.discovery import build
	from apiclient.errors import HttpError
	from oauth2client.tools import argparser

	# Downloading YouTube videos
	import pafy

	<!DOCTYPE html>
	<meta charset="utf-8">

	<!-- The visual styling of our barchart is in this file -->
	<link rel="stylesheet" href="barchart.css">

	<!-- This is the main container to which we append all chart elements -->
	<svg class="chart"></svg>

	<!-- We include D3 from an external location so that we don't have

	import time
	import numpy as np
	import threading

	import tensorflow as tf
	from tensorflow.python.client import timeline


	def test_queue():

	{
	"cmd": ["open", "-a", "Google Chrome", "$file", "--args", "--allow-file-access-from-files"]
	}

	import os
	import sys
	import pickle
	import numpy as np
	import tensorflow as tf
	from tensorflow.python.summary.event_accumulator import EventAccumulator


	event_file = "/home/trunia1/dev/python/LSTMCounting/output/screens/" \
	"summaries/004_rmsprop_0.0005_lstm_512x2_grad_10_batch_128_dropkp_0.75/" \



	# Initialize placeholders for feeding in to the queue
	pl_queue_screens = tf.placeholder(tf.float32, shape=[config.seq_length, config.image_size, config.image_size, config.input_channels], name="queue_inputs")
	pl_queue_targets = tf.placeholder(tf.uint8, shape=[config.seq_length], name="queue_targets_cnt")

	# ...

	capacity = config.min_after_dequeue + 10 * (config.num_gpus*config.batch_size)
	q = tf.RandomShuffleQueue(