-
Notifications
You must be signed in to change notification settings - Fork 1
/
check_pred_vision.py
133 lines (119 loc) · 6.54 KB
/
check_pred_vision.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
import os, cv2
from tqdm import tqdm
import torch, argparse
from pathlib import Path
import numpy as np
from helpers import Helpers
import torch.nn as nn
from vision_pipeline import VISION_PIPELINE, FINAL_DATASET
if __name__ == "__main__":
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
test_folder = 'test_BookShelf_S1'
model_checkpoint = 'vision_checkpointAdam_' + test_folder[5:] + '.pth'
flownet_checkpoint = 'flownets_EPE1.951.pth.tar'
# flownet_checkpoint = 'FlowNet2-SD_checkpoint.pth.tar'
trim_frame_size = 150
pipeline = VISION_PIPELINE(flownet_checkpoint, device)
criterion = nn.L1Loss()
if Path(pipeline.var.root + 'datasets/' + test_folder[5:] + '/' + model_checkpoint).is_file():
checkpoint = torch.load(pipeline.var.root + 'datasets/' + test_folder[5:] + '/' + model_checkpoint)
pipeline.load_state_dict(checkpoint['model_state_dict'])
# pipeline.current_loss = checkpoint['loss']
print('Model loaded')
utils = Helpers(test_folder)
_, _, _, testing_target = utils.load_datasets()
os.chdir(pipeline.var.root)
pipeline.eval()
with torch.no_grad():
testDataset = FINAL_DATASET('testing_images', testing_target)
testLoader = torch.utils.data.DataLoader(testDataset, shuffle=False, batch_size=pipeline.var.batch_size, drop_last=True, num_workers=0)
# tqdm_trainLoader = tqdm(trainLoader)
tqdm_testLoader = tqdm(testLoader)
num_samples = 0
total_loss, total_correct, total_accuracy = [], 0.0, 0.0
predList, labelList, testPD = None, None, None
for batch_index, (feat, labels) in enumerate(tqdm_testLoader):
num_samples += feat.size(0)
labels = labels[:,0,:]
# labels[:,0] *= 0.2667
# labels[:,1] *= 0.3556
pred = pipeline(feat.float()).to(device)
loss = criterion(pred, labels.float())
pred, labels = pipeline.get_original_coordinates(pred, labels)
dist = torch.cdist(pred, labels.float(), p=2)[0].unsqueeze(dim=0)
if batch_index > 0:
testPD = torch.cat((testPD, dist), 1)
predList = torch.cat((predList, pred), 0)
labelList = torch.cat((labelList, labels), 0)
else:
testPD = dist
predList = pred
labelList = labels
total_loss.append(loss.detach().item())
total_correct += pipeline.get_num_correct(pred, labels.float())
total_accuracy = total_correct / num_samples
tqdm_testLoader.set_description('training: ' + '_loss: {:.4} correct: {} accuracy: {:.3} MPD: {}'.format(
np.mean(total_loss), total_correct, 100.0*total_accuracy, torch.mean(testPD)))
os.chdir(pipeline.var.root + test_folder)
video_file = 'scenevideo.mp4'
capture = cv2.VideoCapture(video_file)
frame_count = int(capture.get(cv2.CAP_PROP_FRAME_COUNT))
fps = capture.get(cv2.CAP_PROP_FPS)
# print(frame_count, fps, predList[0], testing_target[0])
capture.set(cv2.CAP_PROP_POS_FRAMES,trim_frame_size+1)
ret, frame = capture.read()
#
# fourcc = cv2.VideoWriter_fourcc(*'MP4V')
# out = cv2.VideoWriter('vision_output.mp4',fourcc, fps, (frame.shape[1],frame.shape[0]))
acc = 0
for i in range(frame_count - 3000):
if ret == True:
# cv2.namedWindow('image', cv2.WINDOW_NORMAL)
# cv2.resizeWindow('image', 512, 512)
# frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
# coordinate = sliced_gaze_dataset[i]
# pred_gaze_pts = coordinate[i]
# for index, pt in enumerate(coordinate):
# try:
# (x, y) = pt[0], pt[1]
# frame = cv2.circle(frame, (int(x*frame.shape[1]),int(y*frame.shape[0])), radius=5, color=(0, 0, 255), thickness=5)
# except Exception as e:
# print(e)
try:
# print(tt[i], testing_target[i])
# frame = cv2.resize(frame, (512, 384))
# gt_gaze_pts = tt[i][0]
pred_gaze_pts = predList[i]
gt_gaze_pts = labelList[i]
# gt_gaze_pts[0] *= 1920.0
# gt_gaze_pts[1] *= 1080.0
# gt_gaze_pts[0] *= 512.0
# gt_gaze_pts[1] *= 384.0
print(gt_gaze_pts, labelList[i])
pred_gaze_pts = predList[i]
# frame = cv2.resize(frame, (512, 384))
padding_r = 50.0
padding = 50.0
sign = 1 if random.random() > 0.5 else -1
# start_point = (int(gt_gaze_pts[0]*frame.shape[1]) - int(padding), int(gt_gaze_pts[1]*frame.shape[0]) + int(padding_r))
# end_point = (int(gt_gaze_pts[0]*frame.shape[1]) + int(padding), int(gt_gaze_pts[1]*frame.shape[0]) - int(padding_r))
# pred_start_point = (int(gt_gaze_pts[0]*frame.shape[1] - sign*padding) - int(padding), int(gt_gaze_pts[1]*frame.shape[0] - sign*padding_r) + int(padding_r))
# pred_end_point = (int(gt_gaze_pts[0]*frame.shape[1] - sign*padding) + int(padding), int(gt_gaze_pts[1]*frame.shape[0] - sign*padding_r) - int(padding_r))
# #
# frame = cv2.rectangle(frame, start_point, end_point, color=(0, 0, 255), thickness=5)
# frame = cv2.rectangle(frame, pred_start_point, pred_end_point, color=(0, 255, 0), thickness=5)
#
# frame = cv2.circle(frame, (int(gt_gaze_pts[0]*frame.shape[1]) ,int(gt_gaze_pts[1]*frame.shape[0])), radius=5, color=(0, 0, 255), thickness=5)
# frame = cv2.circle(frame, (int(gt_gaze_pts[0]*frame.shape[1] - sign*padding) ,int(gt_gaze_pts[1]*frame.shape[0] - sign*padding_r)), radius=5, color=(0, 255, 0), thickness=5)
frame = cv2.circle(frame, (int(gt_gaze_pts[0]),int(gt_gaze_pts[1])), radius=5, color=(0, 0, 255), thickness=5)
frame = cv2.circle(frame, (int(pred_gaze_pts[0]),int(pred_gaze_pts[1])), radius=5, color=(0, 255, 0), thickness=5)
# correct = torch.logical_and((torch.abs(pred_gaze_pts[0] - gt_gaze_pts[0]) <= 100.0), (torch.abs(pred_gaze_pts[1]-gt_gaze_pts[1]) <= 100.0)).sum().item()
# print(pred_gaze_pts, gt_gaze_pts, correct)
except Exception as e:
print(e)
cv2.imshow('image', frame)
# out.write(frame)
# if cv2.waitKey(1) & 0xFF == ord('q'):
# break
cv2.waitKey(0)
ret, frame = capture.read()