基于LK方法的实时光流计算
这个示例展示一个在AOP上运行的LK方法
调用接口： - tianmoucv.proc.opticalflow.LK_optical_flow
%load_ext autoreload
The autoreload extension is already loaded. To reload it, use:
%reload_ext autoreload
引入必要的包
%autoreload
import sys
import numpy as np
import os
import cv2
import matplotlib.pyplot as plt
import torch
import torch.nn.functional as F
import math
from tianmoucv.data import TianmoucDataReader
TianMouCV™ 0.3.5.4, via Y. Lin update new nn for reconstruction
数据构造
train='/data/lyh/tianmoucData/tianmoucReconDataset/train/'
dirlist = os.listdir(train)
traindata = [train + e for e in dirlist]
val='/data/lyh/tianmoucData/tianmoucReconDataset/test/'
vallist = os.listdir(val)
valdata = [val + e for e in vallist]
key_list = []
print('---------------------------------------------------')
for sampleset in traindata:
print('---->',sampleset,'有：',len(os.listdir(sampleset)),'个样本')
for e in os.listdir(sampleset):
print(e,end=" ")
key_list.append(e)
print('---------------------------------------------------')
for sampleset in valdata:
print('---->',sampleset,'有：',len(os.listdir(sampleset)),'个样本')
for e in os.listdir(sampleset):
print(e,end=" ")
key_list.append(e)
all_data = valdata + traindata
---------------------------------------------------
----> /data/lyh/tianmoucData/tianmoucReconDataset/train/normal 有： 67 个样本
outdoor_cross_8 train_cross2 traffic5 indoor_office_2 train_indoor_dog4 outdoor_cross_5 indoor_office_6 train_running_man_5 indoor_office_1 train_exam_fan2 indoor_office_3 people1 train_exam_fan5 indoor_office_4 indoor_slefie_2 outdoor_cross_9 outdoor_bridge_1 outdoor_cross_4 outdoor_cross_1 outdoor_4huan traffic15 outdoor_cross_12 outdoor_bridge_2 traffic9 traffic2 traffic_nohdr_16 traffic11 train_exam_fan1 train_indoor_dog1 train_cross3 train_driving5 traffic7 traffic_nohdr_15 train_driving14 train_driving9 outdoor_cross_7 train_driving4 traffic10 train_running_man_6 train_exam_fan3 train_driving6 train_cross4 train_driving3 outdoor_cross_3 train_driving11 traffic14 outdoor_bz_1 outdoor_hutong_1 indoor_slefie_1 indoor_keyboard1 train_man_play_ball1 train_driving8 traffic3 train_driving7 outdoor_cross_11 train_exam_full4 train_running_man_7 people10 traffic6 train_driving13 traffic13 traffic_nohdr_17 train_driving10 train_exam_full2 train_indoor_dog2 traffic1 train_exam_full1 ----> /data/lyh/tianmoucData/tianmoucReconDataset/train/extreme 有： 51 个样本
flicker_12 underbridge_hdr_4 hdr_people9 train_exam_flicker3 underbridge_hdr_2 hdr_traffic35 hdr_people15 flicker_3 hdr_people2 train_tunnel3_hdr_ae hdr_traffic18 shake2 indoor_crazy_shake flicker_1 flicker_8 hdr_traffic20 underbridge_hdr_1 hdr_traffic30 train_exam_flicker2 hdr_traffic19 flicker_17 flicker_6 shake5 hdr_traffic23 train_exam_flicker1 train_hdr_human hdr_people5 hdr_people3 flicker_0 hdr_people11 train_tunnel6_hdr_ae flicker_4 flicker_9 flicker_11 flicker_15 hdr_people7 shake4 hdr_traffic26 train_tunnel4_hdr_ae hdr_traffic25 hdr_traffic29 train_tunnel1_hdr_blur shake1 train_driving2 hdr_traffic22 train_exam_fan_QRcode_1 hdr_people6 flicker_14 hdr_traffic34 hdr_people14 train_tunnel5_hdr_ae ---------------------------------------------------
----> /data/lyh/tianmoucData/tianmoucReconDataset/test/normal 有： 24 个样本
test_tunnel2 test_man_play_ball3 test_exam_fan4 test_driving24 test_driving3 test_driving20 indoor_office_5 outdoor_cross_10 test_running_man_8 test_cross3 outdoor_cross_13 outdoor_4huan_2 test_exam_full3 test_driving4 traffic4 test_driving12 test_driving16 outdoor_cross_6 traffic8 test_driving8 traffic12 outdoor_bridge_3 test_running_man_4 indoor_keyboard2 ----> /data/lyh/tianmoucData/tianmoucReconDataset/test/extreme 有： 30 个样本
shake3 test_tunnel7_hdr_ae hdr_traffic36 test_exam_fan_QRcode_2 flicker_16 hdr_traffic21 hdr_traffic32 test_indoor_dog3 hdr_traffic24 train_exam_flicker5 hdr_people13 test_tunnel8_hdr_ae_double hdr_people8 flicker_13 hdr_traffic33 hdr_people4 test_exam_fan_QRcode_3 hdr_traffic31 indoor_selfie_shake_3 flicker_7 hdr_people16 flicker_10 flicker_2 hdr_people12 test_driving_night_light1 test_hdr_human2 underbridge_hdr_3 flicker_18 flicker_5 shake6
光流计算
%autoreload
from tianmoucv.proc.opticalflow import interpolate_image,flow_to_image
from tianmoucv.proc.opticalflow import LK_optical_flow
from tianmoucv.isp import SD2XY
import time
from IPython.display import clear_output
accumTime = 5
noiseThresh = 8
lambda_of_HS = 25 #bigger->smoother
#(输入是0~255时lambda要>1,否则千万不能太大)
W = 640
H = 320
gridX, gridY = np.meshgrid(np.arange(W), np.arange(H))
key_list = ['test_exam_fan4']
imlist = []
for key in key_list:
pathList = all_data
dataset = TianmoucDataReader(pathList,showList=True,
matchkey = key,
MAXLEN=-1,
speedUpRate=1,
print_info=False)
for index in range(len(dataset)):
if index <= 15:
continue
elif index > 20:
break
else:
print('rpogress:',index,'/',len(dataset))
sample = dataset[index]
F0 = sample['F0']
F1 = sample['F1']
tsdiff = sample['rawDiff']
F0show = F0.copy()
show_img = F0show.copy()
for b in range(25//accumTime):
sd = 0
td = 0
TD = 0
#积累几帧diff
for t in range(accumTime):
threshed_tsdiff = tsdiff[:,b*accumTime+t,...].permute(1,2,0)
threshed_tsdiff[abs(threshed_tsdiff)<noiseThresh] = 0
SD = threshed_tsdiff[...,1:]
TD = threshed_tsdiff[...,0]
Ix,Iy= SD2XY(SD)
sd += torch.FloatTensor(np.stack([Ix,Iy],axis=0))
td += -(TD)
# AOP预处理
sd = sd/accumTime
td = td.unsqueeze(0)
td = F.interpolate(td.unsqueeze(0), size=sd.shape[1:], mode='bilinear').squeeze(0)
# 计算OF
rawflow = LK_optical_flow(sd,td, win=31, stride = 5,mask=None,ifInterploted = False)
u = rawflow[0,:, :].numpy()
v = rawflow[1,:, :].numpy()
u = torch.Tensor(cv2.resize(u,(640,320))).unsqueeze(0)
v = torch.Tensor(cv2.resize(v,(640,320))).unsqueeze(0)
#可视化
flow_show = flow_to_image(rawflow.permute(1,2,0).numpy())
flow_show = torch.Tensor(cv2.resize(flow_show,(640,320)))/255.0
flow_show = (flow_show*255).numpy().astype(np.uint8)
tdshow = TD.unsqueeze(0).unsqueeze(0)
tdshow = F.interpolate(tdshow,(320,640),mode='bilinear')
mask = np.mean(flow_show,axis=-1) > 225
flow_show[np.stack([mask]*3,axis=-1)]=0
show_img = interpolate_image(show_img,u,v)
tdiff_show = np.stack([tdshow[0,0,...].cpu()*255]*3,axis=2).astype(np.uint8)
sparsity = 8
scale = 10
for w in range(640//sparsity):
for h in range(320//sparsity):
x = int(w*sparsity)
y = int(h*sparsity)
u_ij = -u[0,y,x]
v_ij = -v[0,y,x]
color = flow_show[y,x,:]
color = tuple([int(e+20) for e in color])
if (u_ij**2+v_ij**2)>5:
cv2.arrowedLine(flow_show, (x,y), (int(x+u_ij*scale),int(y+v_ij*scale)), color,2, tipLength=0.15)
tdiff_show_tensor = torch.Tensor(tdiff_show.copy())
flow_show_tensor = torch.Tensor(flow_show)
mask = torch.stack([torch.mean(flow_show_tensor,dim=-1)>0]*3,dim=-1)
tdiff_show_tensor[mask] = flow_show_tensor[mask]/255.0
tdiff_show_merge = tdiff_show_tensor.numpy()
imshow = np.concatenate([flow_show/255.0,tdiff_show,tdiff_show_merge],axis=0)
imlist.append(imshow)
if b %10 ==0:
clear_output()
plt.figure(figsize=(9,5))
plt.axis('off')
plt.subplot(2,2,1)
plt.imshow(Ix,cmap='gray')
plt.subplot(2,2,2)
plt.imshow(TD,cmap='gray')
plt.axis('off')
plt.subplot(2,2,3)
plt.imshow(F0show)
plt.subplot(2,2,4)
plt.imshow(flow_show)
plt.show()
def images_to_video(frame_list,name,Val_size=(512,256),Flip=False):
fps = 30
size = (Val_size[0], Val_size[1]) # 需要转为视频的图片的尺寸
out = cv2.VideoWriter(name,0x7634706d , fps, size)
for frame in frame_list:
frame = (frame-np.min(frame))/(np.max(frame)-np.min(frame)) * 255
frame2 = frame.astype(np.uint8)
out.write(frame2)
out.release()
images_to_video(imlist,'./LKOF_'+key+'.mp4',Val_size=(640,960),Flip=True)