forked from Alibaba-MIIL/ML_Decoder
-
Notifications
You must be signed in to change notification settings - Fork 0
/
infer_2.py
163 lines (126 loc) · 6.12 KB
/
infer_2.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
import os
import argparse
import time
import torch
import torch.nn.parallel
import torch.optim
import torch.utils.data.distributed
from src_files.helper_functions.bn_fusion import fuse_bn_recursively
from src_files.models import create_model
import matplotlib
from src_files.models.tresnet.tresnet import InplacABN_to_ABN
matplotlib.use('TkAgg')
matplotlib.use('TkAgg')
import matplotlib.pyplot as plt
from PIL import Image
import numpy as np
import tools
import pandas as pd
import cv2
parser = argparse.ArgumentParser(description='PyTorch MS_COCO infer')
parser.add_argument('--num-classes', default=80, type=int)
parser.add_argument('--model-path', type=str, default='./models_local/TRresNet_L_448_86.6.pth')
parser.add_argument('--pic-path', type=str, default='./pics/000000000885.jpg')
parser.add_argument('--model-name', type=str, default='tresnet_l')
parser.add_argument('--image-size', type=int, default=448)
# parser.add_argument('--dataset-type', type=str, default='MS-COCO')
parser.add_argument('--th', type=float, default=0.75)
parser.add_argument('--top-k', type=float, default=100)
# ML-Decoder
parser.add_argument('--use-ml-decoder', default=1, type=int)
parser.add_argument('--num-of-groups', default=-1, type=int) # full-decoding
parser.add_argument('--decoder-embedding', default=768, type=int)
parser.add_argument('--zsl', default=0, type=int)
def main():
print('Generating detections...')
# parsing args
args = parser.parse_args()
# Setup model from checkpoint
if 'ckpt' in args.model_path:
mode = 'ckpt'
checkpoint = torch.load(args.model_path)
model = create_model(args, load_head=True, from_ckpt=True).cuda()
model.load_state_dict(checkpoint)
model.eval()
else:
# Setup model
mode = 'pt'
print('creating model {}...'.format(args.model_name))
model = create_model(args, load_head=True).cuda()
state = torch.load(args.model_path, map_location='cpu')
model.load_state_dict(state['model'], strict=True)
########### eliminate BN for faster inference ###########
model = model.cpu()
model = InplacABN_to_ABN(model)
model = fuse_bn_recursively(model)
model = model.cuda().half().eval()
#######################################################
print('model has been loaded.')
root = '/home/brandon/Documents/datasets/mscoco'
ds_split = 'test'
print('loading ' + ds_split + ' dataset...')
ds = tools.KarpathySplits(root, ds_split)
print(len(ds), 'images were loaded.')
if mode == 'ckpt':
classes_list = np.array(ds.keywords)
else:
classes_list = np.array(list(state['idx_to_class'].values()))
print('loading keywords intersected with mscoco..')
classes_mscoco_path = '/home/brandon/Documents/git/phd/phd-image-captioning/keywords-predictor/openimages/mscoco_full_intersected_openimages_dist.csv'
classes_mscoco = pd.read_csv(classes_mscoco_path)['Description'].values
print('mscoco_classes:', len(classes_mscoco))
columns_list = ['img_path', 'actual_classes', 'pred_classes', 'num_actual_classes', 'num_pred_classes', 'pred_scores']
predictions = pd.DataFrame(columns=columns_list)
max_imgs = len(ds)
for i, img in enumerate(ds):
if i < max_imgs:
target = img[1]
img_path = img[2]
print(i, img_path)
# doing inference
im = Image.open(img_path)
im_resize = im.resize((args.image_size, args.image_size))
np_img = np.array(im_resize, dtype=np.uint8)
if len(np_img.shape) == 2:
np_img = cv2.cvtColor(np_img, cv2.COLOR_GRAY2RGB)
tensor_img = torch.from_numpy(np_img).permute(2, 0, 1).float() / 255.0 # HWC to CHW
tensor_batch = torch.unsqueeze(tensor_img, 0).cuda().half() # float16 inference
output = torch.squeeze(torch.sigmoid(model(tensor_batch)))
np_output = output.cpu().detach().numpy()
## Top-k predictions
idx_th = np_output > args.th
detected_classes = classes_list[idx_th]
scores = np_output[idx_th]
detected_mscoco_classes = []
scores_mscoco = []
for i, d in enumerate(detected_classes):
d = d.lower()
if d in classes_mscoco:
detected_mscoco_classes += [d]
scores_mscoco += [scores[i]]
#print(detected_mscoco_classes)
#print(scores_mscoco)
scores_idx = np.argsort(-np.array(scores_mscoco))
#print(np.array(scores_mscoco)[scores_idx])
actual_classes = tools.convert_target_to_keywords(target, ds.keywords)
detected_classes = list(np.array(detected_mscoco_classes)[scores_idx][:args.top_k])
predictions = predictions.append(pd.DataFrame([[img_path, actual_classes, detected_classes, len(actual_classes), len(detected_classes), scores_mscoco]], columns=columns_list))
# displaying image
# print('showing image on screen...')
# fig = plt.figure()
# plt.imshow(im)
# plt.axis('off')
# plt.axis('tight')
# plt.rcParams["axes.titlesize"] = 10
# plt.title("detected classes: {}".format(detected_classes))
# plt.show()
# print('done\n')
else:
break
predictions['idx'] = range(0, len(predictions)-1)
predictions.to_csv('ml_predictions_' + ds_split + '.csv', index=False)
if __name__ == '__main__':
main()
#python infer_2.py --num-classes=9605 --model-name=tresnet_m --model-path=./models_zoo/tresnet_m_open_images_200_groups_86_8.pth --num-of-groups=200 --image-size=224
#python infer_2.py --num-classes=1000 --model-name=tresnet_m --model-path=./models/openimages_b_64_labels_1000_g_4_lr_2e-4_train_decoder_grps_200/model-highest.ckpt --num-of-groups=200 --image-size=224
#python infer_2.py --num-classes=3940 --model-name=tresnet_m --model-path=./models/openimages_b_64_labels_3940_g_4_lr_2e-4_train_decoder_grps_200/model-highest.ckpt --num-of-groups=200 --image-size=224