-
Notifications
You must be signed in to change notification settings - Fork 1
/
demo.py
176 lines (152 loc) · 7.45 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
import importlib
import tkinter
import imageio
import torch
import cv2
from PIL import Image, ImageTk
from objects import *
from utils.video import *
from utils.config import Config
from utils.formatting import *
from tasks.lane_detection import LaneDetector
from tasks.e2e import *
config = Config()
Net = importlib.import_module(config['model']['pyPath']).Net
steer_avg, steer_std = None, None
motor_avg, motor_std = None, None
accel_avg, accel_std = None, None
torch.cuda.set_device(config['hardware']['gpu'])
display_dims = (1280, 720)
resolution = (1280, 720)
fps_downsample = 6
downsample_factor = 1./4.
downsampled_resolution = (round(resolution[0] * downsample_factor), round(resolution[1] * downsample_factor))
input_frames = config['constants']['inputFrames']
demo_lenth_factor = 1.
class Demo:
"""
Demo application, displays video output.
"""
def __init__(self, beam_path):
"""
:param beam_path: <str> Path to beam to demo
"""
self.beam = E2EBeam(filepath=beam_path, demo=True)
self.bricks = self.beam.get_bricks(sort_by=lambda brick: brick.frame)
self.vid_path = self.beam.source_path
self.vid_reader = imageio.get_reader(self.vid_path, 'ffmpeg')
self.vid_writer = imageio.get_writer('./demo_normal.mp4', fps=self.vid_reader.get_meta_data()['fps'])
self.vid_frames = get_vid_frames(self.vid_reader,
num_frames=round(self.beam.num_frames * fps_downsample),
resolution=resolution,
downsample_factor=1.,
fps_downsample=1)
self.downsampled_vid_frames = get_vid_frames(num_frames=round(self.beam.num_frames * fps_downsample * demo_lenth_factor),
vid_frames=self.vid_frames,
resolution=resolution,
downsample_factor=downsample_factor,
fps_downsample=1)
self.fps = self.vid_reader.get_meta_data()['fps']
self.controls = []
self.n_frames_feed = config['constants']['inputFrames']
self.lane_detector = LaneDetector()
# Load network
self.net = Net(config=config).cuda()
self.net.load(config['model']['saveDirPath'], config['model']['name'], 'core', config['demo']['epoch'])
self.task = E2ETask(name='e2e', config=config, load_dataset=False)
self.net.eval()
self.task.eval()
self.task.net.load(config['model']['saveDirPath'], config['model']['name'], self.task.name, config['demo']['epoch'])
self.task.cuda()
self.calc_controls()
self.curr_img, self.imgtk = None, None
self.curr_frame_i = 0
# Set up tkinter
self.root = tkinter.Tk()
self.root.title('Demo')
self.root.geometry(str(display_dims[0]) + 'x' + str(display_dims[1]))
self.canvas = tkinter.Canvas(self.root, width=display_dims[0], height=display_dims[1])
self.canvas.place(x=0, y=0)
self.canvas_img = None
self.overlay_line = self.canvas.create_line(display_dims[0] // 2,
display_dims[1] // 2,
display_dims[0] // 2,
display_dims[1] // 2,
fill='red',
arrow='last')
self.root.config(cursor="arrow")
self.display_loop()
def get_video(self, frame_i):
if frame_i < (self.n_frames_feed - 1) * fps_downsample:
return None
else:
vid_frames = get_vid_frames(num_frames=input_frames,
vid_frames=self.downsampled_vid_frames,
resolution=downsampled_resolution,
downsample_factor=1.,
fps_downsample=fps_downsample,
start_frame=frame_i - ((self.n_frames_feed - 1) * fps_downsample))
return vid_frames
def get_metadata(self, frame_i):
for i in range(len(self.bricks) - 1):
curr_brick, next_brick = self.bricks[i], self.bricks[i + 1]
if curr_brick.frame * fps_downsample <= frame_i < next_brick.frame * fps_downsample:
return torch.LongTensor(np.array([0]))
return torch.LongTensor(np.array([0]))
def calc_controls(self):
for frame_i in range(round(36 * self.fps * demo_lenth_factor)):
vid_frames = self.get_video(frame_i)
if frame_i % 100 == 0:
print(frame_i)
if vid_frames is None:
self.controls.append((0, 0, 0))
else:
metadata = torch.unsqueeze(self.get_metadata(frame_i).cuda(), 0)
vid_frames = torch.unsqueeze(torch.FloatTensor(vid_frames).cuda(), 0)
embedding = self.net(vid_frames, metadata)
output = self.task.net(embedding).data[0]
controls = (output[0], output[1], output[2])
self.controls.append(controls)
def display_loop(self):
curr_control = self.get_control(self.curr_frame_i)
self.curr_img = self.vid_frames[self.curr_frame_i]
self.curr_img = Image.fromarray(
self.paint_overlay_line(
self.lane_detector.detect(
self.curr_img, roi_only=True),
curr_control[0], curr_control[1]))
imgtk = ImageTk.PhotoImage(image=self.curr_img)
self.imgtk = imgtk
if self.canvas_img is not None:
self.canvas.delete(self.canvas_img)
self.canvas_img = self.canvas.create_image(display_dims[0] // 2, display_dims[1] // 2, image=self.imgtk)
# self.calc_overlay_line(curr_control[0], curr_control[1])
self.root.after(17, self.display_loop)
self.curr_frame_i = (self.curr_frame_i + 1) % round(self.beam.num_frames * fps_downsample * demo_lenth_factor)
self.vid_writer.append_data(np.array(np.array(self.curr_img)))
def calc_overlay_line(self, steer, accel):
x_0, y_0 = display_dims[0] // 2, display_dims[1] // 2
x_1, y_1 = x_0 + steer * 25, y_0 - round((accel * 1.14) * 200)
self.canvas.delete(self.overlay_line)
self.overlay_line = self.canvas.create_line(x_0, y_0, x_1, y_1, fill='red', arrow='last')
def paint_overlay_line(self, image, steer, accel):
print(steer, accel)
x_0, y_0 = image.shape[1] // 2, image.shape[0] // 2
x_1, y_1 = round(x_0 + steer * 50), y_0 - round((accel * 1.14) * 200)
cv2.arrowedLine(image, (x_0, y_0), (x_1, y_1), (255, 0, 0), 3)
return image
def get_control(self, frame_i):
control = [0, 0, 0]
total_weight = 0
for i in range(14, -1, -1):
curr_i = frame_i - i
curr_control = self.controls[curr_i]
curr_weight = 1. / (i + 1.) ** 0.5
control[0] += curr_weight * curr_control[0]
control[1] += curr_weight * curr_control[1]
control[2] += curr_weight * curr_control[2]
total_weight += curr_weight
control = [_/total_weight for _ in control]
return control
app = Demo('/data/datasets/processed/beams/294.beam')
app.root.mainloop()