-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathinfer.py
147 lines (123 loc) · 5.85 KB
/
infer.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
import torch
import argparse
import os
import matplotlib.pyplot as plt
import torchvision.utils
from PIL import Image
import numpy as np
from models.models import MTResnetAggregate
import json
import time
# ----------------------------------------------------------------------
# Parameters
parser = argparse.ArgumentParser(
description='PETA: Photo album Event recognition using Transformers Attention.')
parser.add_argument('--model_path', type=str,
default='/content/drive2/checkpoints/event_importance_tresnet/version_2/checkpoints/best-epoch=119-mAP=67.21.ckpt')
parser.add_argument('--album_path', type=str,
default='../CUFED_split/images/test/61_33988590@N00')
parser.add_argument('--image_importance_pth', type=str,
default='../CUFED_split/image_importance.json')
parser.add_argument('--event_type_pth', type=str,
default='../CUFED_split/event_type.json')
parser.add_argument('---album_list', type=str,
default='filenames/test.txt')
parser.add_argument('--val_dir', type=str, default='./albums')
parser.add_argument('--num_classes', type=int, default=23)
parser.add_argument('--model_name', type=str, default='mtresnetaggregate')
parser.add_argument('--transformers_pos', type=int, default=0)
parser.add_argument('--input_size', type=int, default=224)
parser.add_argument('--transform_type', type=str, default='squish')
parser.add_argument('--backbone', type=str, default='tresnet_m')
parser.add_argument('--album_sample', type=str, default='rand_permute')
parser.add_argument('--dataset_path', type=str, default='./data/ML_CUFED')
parser.add_argument('--dataset_type', type=str, default='ML_CUFED')
parser.add_argument('--path_output', type=str, default='./outputs')
parser.add_argument('--use_transformer', type=int, default=1)
parser.add_argument('--infer', type=int, default=1)
parser.add_argument('--album_clip_length', type=int, default=32)
parser.add_argument('--batch_size', type=int, default=32)
parser.add_argument('--num_workers', type=int, default=0)
parser.add_argument('--top_k', type=int, default=3)
parser.add_argument('--threshold', type=float, default=0.5)
parser.add_argument('--remove_model_jit', type=int, default=None)
parser.add_argument('--attention', type=str,
default='multihead')
def get_album(args):
files = os.listdir(args.album_path)
album_name = args.album_path.rsplit('/', 1)[1]
cls_dict = json.load(open(args.event_type_pth))
event = cls_dict[album_name]
n_files = len(files)
idx_fetch = np.linspace(0, n_files-1, args.album_clip_length, dtype=int)
tensor_batch = torch.zeros(
len(idx_fetch), args.input_size, args.input_size, 3)
for i, id in enumerate(idx_fetch):
im = Image.open(os.path.join(args.album_path, files[id]))
im_resize = im.resize((args.input_size, args.input_size))
np_img = np.array(im_resize, dtype=np.uint8)
tensor_batch[i] = torch.from_numpy(np_img).float() / 255.0
tensor_batch = tensor_batch.permute(0, 3, 1, 2).cuda() # HWC to CHW
# tensor_images = torch.unsqueeze(tensor_images, 0).cuda()
montage = torchvision.utils.make_grid(tensor_batch).permute(1, 2, 0).cpu()
return tensor_batch, montage, event
def inference(tensor_batch, model, classes_list, args):
output, _ = model(tensor_batch)
output = torch.squeeze(torch.sigmoid(output))
np_output = output.cpu().detach().numpy()
idx_sort = np.argsort(-np_output)
# Top-k
detected_classes = np.array(classes_list)[idx_sort][: args.top_k]
scores = np_output[idx_sort][: args.top_k]
# Threshold
idx_th = scores > args.threshold
return detected_classes[idx_th], scores[idx_th]
def display_image(im, tags, event, filename, path_dest):
if not os.path.exists(path_dest):
os.makedirs(path_dest)
plt.figure()
plt.imshow(im)
plt.axis('off')
plt.axis('tight')
plt.rcParams["axes.titlesize"] = 16
plt.title("Predicted classes: {} \n True classes: {}".format(tags, event))
plt.savefig(os.path.join(path_dest, filename))
def load_model(net, path):
if path is not None and path.endswith(".ckpt"):
state_dict = torch.load(path, map_location='cpu')
if "state_dict" in state_dict:
state_dict = state_dict["state_dict"]
compatible_state_dict = {}
for k, v in state_dict.items():
if k.startswith('net.'):
compatible_state_dict[k.replace('net.', '')] = v
net.load_state_dict(compatible_state_dict)
return net
def main(classes_list):
start_time = time.time()
args = parser.parse_args()
# net = EventNetwork(encoder_name='resnet101', num_classes=args.num_classes).cuda()
# net.eval()
net = MTResnetAggregate(args).cuda()
net.eval()
net = load_model(net, args.model_path)
# Get album
tensor_batch, montage, event = get_album(args)
# Inference
tags, confs = inference(tensor_batch, net, classes_list, args)
print("--- %s seconds ---" % (time.time() - start_time))
# Visualization
display_image(montage, tags, event, 'result.jpg', os.path.join(
args.path_output, args.album_path.rsplit('/')[1]).replace("./albums", ""))
# Actual validation process
# print('loading album and doing inference...')
# map = validate(model, val_loader, classes_list, args.threshold)
# print("final validation map: {:.2f}".format(map))
print('Done\n')
if __name__ == '__main__':
classes_list = ['Architecture', 'BeachTrip', 'Birthday', 'BusinessActivity',
'CasualFamilyGather', 'Christmas', 'Cruise', 'Graduation', 'GroupActivity',
'Halloween', 'Museum', 'NatureTrip', 'PersonalArtActivity',
'PersonalMusicActivity', 'PersonalSports', 'Protest', 'ReligiousActivity',
'Show', 'Sports', 'ThemePark', 'UrbanTrip', 'Wedding', 'Zoo']
main(classes_list)