-
Notifications
You must be signed in to change notification settings - Fork 8
/
demo.py
180 lines (156 loc) · 8.81 KB
/
demo.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
# -*- coding: utf-8 -*-
"""
Demo for histogram layer networks (HistRes_B)
Current script is only for experiments on
single cpu/gpu. If you want to run the demo
on multiple gpus (two were used in paper),
please contact me at jpeeples@ufl.edu
for the parallel version of
demo.py
@author: jpeeples
"""
## Python standard libraries
from __future__ import print_function
from __future__ import division
import numpy as np
import os
## PyTorch dependencies
import torch
import torch.nn as nn
import torch.optim as optim
## Local external libraries
from Utils.Network_functions import initialize_model, train_model,test_model
from Utils.RBFHistogramPooling import HistogramLayer as RBFHist
from Utils.LinearHistogramPooling import HistogramLayer as LinearHist
from Utils.Save_Results import save_results
from Demo_Parameters import Network_parameters
from Prepare_Data import Prepare_DataLoaders
#Name of dataset
Dataset = Network_parameters['Dataset']
#Model(s) to be used
model_name = Network_parameters['Model_names'][Dataset]
#Number of classes in dataset
num_classes = Network_parameters['num_classes'][Dataset]
#Number of runs and/or splits for dataset
numRuns = Network_parameters['Splits'][Dataset]
#Number of bins and input convolution feature maps after channel-wise pooling
numBins = Network_parameters['numBins']
num_feature_maps = Network_parameters['out_channels'][model_name]
#Local area of feature map after histogram layer
feat_map_size = Network_parameters['feat_map_size']
# Detect if we have a GPU available
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
#Location to store trained models
current_directory = os.getcwd()
final_directory = os.path.join(current_directory, Network_parameters['folder'])
print('Starting Experiments...')
for split in range(0, numRuns):
#Keep track of the bins and widths as these values are updated each
#epoch
saved_bins = np.zeros((Network_parameters['num_epochs']+1,
numBins*int(num_feature_maps/(feat_map_size*numBins))))
saved_widths = np.zeros((Network_parameters['num_epochs']+1,
numBins*int(num_feature_maps/(feat_map_size*numBins))))
#Initialize histogram layer based on type
if Network_parameters['histogram_type'] == 'RBF':
histogram_layer = RBFHist(int(num_feature_maps/(feat_map_size*numBins)),
Network_parameters['kernel_size'][model_name],
num_bins=numBins,stride=Network_parameters['stride'],
normalize_count=Network_parameters['normalize_count'],
normalize_bins=Network_parameters['normalize_bins'])
elif Network_parameters['histogram_type'] == 'Linear':
histogram_layer = LinearHist(int(num_feature_maps/(feat_map_size*numBins)),
Network_parameters['kernel_size'][model_name],
num_bins=numBins,stride=Network_parameters['stride'],
normalize_count=Network_parameters['normalize_count'],
normalize_bins=Network_parameters['normalize_bins'])
else:
raise RuntimeError('Invalid type for histogram layer')
# Initialize the histogram model for this run
model_ft, input_size = initialize_model(model_name, num_classes,
Network_parameters['in_channels'][model_name],
num_feature_maps,
feature_extract = Network_parameters['feature_extraction'],
histogram= Network_parameters['histogram'],
histogram_layer=histogram_layer,
parallel=Network_parameters['parallel'],
use_pretrained=Network_parameters['use_pretrained'],
add_bn = Network_parameters['add_bn'],
scale = Network_parameters['scale'],
feat_map_size=feat_map_size)
# Send the model to GPU if available
model_ft = model_ft.to(device)
#Print number of trainable parameters
num_params = sum(p.numel() for p in model_ft.parameters() if p.requires_grad)
print("Number of parameters: %d" % (num_params))
print("Initializing Datasets and Dataloaders...")
# Create training, validation, and test dataloaders
dataloaders_dict = Prepare_DataLoaders(Network_parameters,split,input_size=input_size)
#Save the initial values for bins and widths of histogram layer
#Set optimizer for model
if(Network_parameters['histogram']):
reduced_dim = int((num_feature_maps/feat_map_size)/(numBins))
if (Network_parameters['in_channels'][model_name]==reduced_dim):
dim_reduced = False
saved_bins[0,:] = model_ft.histogram_layer.centers.detach().cpu().numpy()
saved_widths[0,:] = model_ft.histogram_layer.widths.reshape(-1).detach().cpu().numpy()
else:
dim_reduced = True
saved_bins[0,:] = model_ft.histogram_layer[-1].centers.detach().cpu().numpy()
saved_widths[0,:] = model_ft.histogram_layer[-1].widths.reshape(-1).detach().cpu().numpy()
if(Network_parameters['add_bn']):
optimizer_ft = optim.SGD([
{'params': model_ft.backbone.conv1.parameters()},
{'params': model_ft.backbone.bn1.parameters()},
{'params': model_ft.backbone.layer1.parameters()},
{'params': model_ft.backbone.layer2.parameters()},
{'params': model_ft.backbone.layer3.parameters()},
{'params': model_ft.backbone.layer4.parameters()},
{'params': model_ft.histogram_layer.parameters(), 'lr': Network_parameters['new_lr']},
{'params': model_ft.fc.parameters(), 'lr': Network_parameters['new_lr']},
{'params': model_ft.bn_norm.parameters(), 'lr': Network_parameters['new_lr']}
], lr=Network_parameters['pt_lr'], momentum=Network_parameters['momentum'])
else:
optimizer_ft = optim.SGD([
{'params': model_ft.backbone.conv1.parameters()},
{'params': model_ft.backbone.bn1.parameters()},
{'params': model_ft.backbone.layer1.parameters()},
{'params': model_ft.backbone.layer2.parameters()},
{'params': model_ft.backbone.layer3.parameters()},
{'params': model_ft.backbone.layer4.parameters()},
{'params': model_ft.histogram_layer.parameters(), 'lr': Network_parameters['new_lr']},
{'params': model_ft.fc.parameters(), 'lr': Network_parameters['new_lr']},
], lr=Network_parameters['pt_lr'], momentum=Network_parameters['momentum'])
else:
saved_bins = None
saved_widths = None
optimizer_ft = optim.SGD([
{'params': model_ft.conv1.parameters()},
{'params': model_ft.bn1.parameters()},
{'params': model_ft.layer1.parameters()},
{'params': model_ft.layer2.parameters()},
{'params': model_ft.layer3.parameters()},
{'params': model_ft.layer4.parameters()},
{'params': model_ft.fc.parameters(), 'lr': Network_parameters['new_lr']},
], lr=Network_parameters['pt_lr'], momentum = Network_parameters['momentum'])
# Setup the loss fxn
criterion = nn.CrossEntropyLoss()
scheduler = optim.lr_scheduler.StepLR(optimizer_ft,
step_size=Network_parameters['step_size'],
gamma= Network_parameters['gamma'])
# Train and evaluate
train_dict = train_model(
model_ft, dataloaders_dict, criterion, optimizer_ft, device,
saved_bins=saved_bins,saved_widths=saved_widths,histogram=Network_parameters['histogram'],
num_epochs=Network_parameters['num_epochs'],scheduler=scheduler,
dim_reduced=dim_reduced)
test_dict = test_model(dataloaders_dict['test'],model_ft,device)
# Save results
if(Network_parameters['save_results']):
save_results(train_dict,test_dict,split,Network_parameters,num_params)
del train_dict,test_dict
torch.cuda.empty_cache()
if(Network_parameters['histogram']):
print('**********Run ' + str(split + 1) + ' For ' + Network_parameters['hist_model'] + ' Finished**********')
else:
print('**********Run ' + str(split + 1) + ' For GAP_' + model_name + ' Finished**********')