qfgaohao

# Create a simple TF Graph 
# By Omid Alemi - Jan 2017
# Works with TF <r1.0

import tensorflow as tf

I = tf.placeholder(tf.float32, shape=[None,3], name='I') # input
W = tf.Variable(tf.zeros_initializer(shape=[3,2]), dtype=tf.float32, name='W') # weights
b = tf.Variable(tf.zeros_initializer(shape=[2]), dtype=tf.float32, name='b') # biases
O = tf.nn.relu(tf.matmul(I, W) + b, name='O') # activation / output

qfgaohao

import torch
from torchvision import models

def parse(net, inputs = torch.randn(1, 3, 224, 224)):
    with torch.onnx.set_training(net, False):
        trace = torch.onnx.utils._trace(net, inputs)

    graph = trace.graph()
    for n in graph.nodes():
        print(n.scopeName(), n.kind())

qfgaohao

import numpy as np

from caffe2.python import (
    brew,
    model_helper,
    optimizer,
    workspace,
    utils,
)
from caffe2.proto import caffe2_pb2

qfgaohao

#include <iostream>
#include <fstream>
#include <string>
#include <sstream>


#include <google/protobuf/text_format.h>

// for read_from_pbtxt_nocopy
#include <fcntl.h>

qfgaohao

def strip_consts(graph_def, max_const_size=32):
    """Strip large constant values from graph_def."""
    strip_def = tf.GraphDef()
    for n0 in graph_def.node:
        n = strip_def.node.add() 
        n.MergeFrom(n0)
        if n.op == 'Const':
            tensor = n.attr['value'].tensor
            size = len(tensor.tensor_content)
            if size > max_const_size:

qfgaohao

import collections
import numpy as np
import itertools

SSDBoxSizes = collections.namedtuple('SSDBoxSizes', ['min', 'max'])

Spec = collections.namedtuple('Spec', ['feature_map_size', 'shrinkage', 'box_sizes', 'aspect_ratios'])

# the SSD orignal specs
specs = [

qfgaohao

r"""Convert an ImageNet like dataset into tfRecord files, provide a method get_dataset to read the created files.

It has similar functions as ImageFolder in Pytorch.

Modified from
https://raw.githubusercontent.com/tensorflow/models/master/research/slim/datasets/download_and_convert_flowers.py
https://github.com/tensorflow/models/blob/master/research/slim/datasets/flowers.py
"""

from __future__ import absolute_import

qfgaohao

import numpy as np

def compute_average_precision(test_results, num_true_cases):
    """
    It computes average precision based on the definition of Pascal Competition. It computes the under curve area
    of precision and recall. Recall follows the normal definition. Precision is a variant. 
    pascal_precision[i] = typical_precision[i:].max()
    
    Usage:
        test_results = np.array([True, False, True, False, True, False, False, False, False, True])

qfgaohao

from datetime import datetime
import json

class Timer:
    def __init__(self, outputFile):
        self.fout = open(outputFile, 'w')
        self.cache = []
        self.records = {}
    
    def start(self, event="time"):

qfgaohao

#include <stdio.h>

#define N (32)


__global__ void increment(int* time) {
    __shared__ float s[1024];
    for (int i = 0; i < 1024; i++) {
        s[i] = 1.0f;
    }