For PURE, first, I combine the frame images into videos. Then I follow the same pre-processing procedures by using openface to get landmarks and crop faces to save each video into .h5 files, but the performance is poor on PURE data sets.The result is shown below: MAE and RMSE appear slightly normal, but why is the R coefficient so small or even zero?I can make sure that I didn't miswrite my evaluate function, because the same evaluate function works fine on the UBFC-rPPG dataset
![d8f89b577332f576a93224bb53eda12](https://private-user-images.githubusercontent.com/110211332/319112885-2c304a54-55b3-459e-8f1c-edb4d780c15c.png?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3MjM0ODYwMzEsIm5iZiI6MTcyMzQ4NTczMSwicGF0aCI6Ii8xMTAyMTEzMzIvMzE5MTEyODg1LTJjMzA0YTU0LTU1YjMtNDU5ZS04ZjFjLWVkYjRkNzgwYzE1Yy5wbmc_WC1BbXotQWxnb3JpdGhtPUFXUzQtSE1BQy1TSEEyNTYmWC1BbXotQ3JlZGVudGlhbD1BS0lBVkNPRFlMU0E1M1BRSzRaQSUyRjIwMjQwODEyJTJGdXMtZWFzdC0xJTJGczMlMkZhd3M0X3JlcXVlc3QmWC1BbXotRGF0ZT0yMDI0MDgxMlQxODAyMTFaJlgtQW16LUV4cGlyZXM9MzAwJlgtQW16LVNpZ25hdHVyZT04YTc3OTk0NDc2YmIwY2I1NWU5Y2ZjMGU1ODZmMWNiMzM3NTFmM2Y0NDFjZTM2YWMwYTRmOTQ5ODQwMGRlYjE2JlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCZhY3Rvcl9pZD0wJmtleV9pZD0wJnJlcG9faWQ9MCJ9.jhQHBPR7QLzKCAq57PO-oJYBVZx6YSfRucamYlJ4A34)
Here is my test.py file:
import numpy as np
import h5py
from torch import nn
import torch
from PhysNetModel import PhysNet
from utils_data import *
from utils_sig import *
from sacred import Experiment
from sacred.observers import FileStorageObserver
import json
from scipy.stats import pearsonr
from ViT import VideoViT
from ViT2 import TransformerImageResizer
from UniFormer import Uniformer
from CNNandTransformer import MyNet
from scipy import interpolate
import json
ex = Experiment('model_pred', save_git_info=False)
@ex.config
def my_config():
e = 29 # the model checkpoint at epoch e
train_exp_num = 3 # the training experiment number
train_exp_dir = './results4cnn_pure/%d'%train_exp_num # training experiment directory
time_interval = 10 # get rppg for 30s video clips, too long clips might cause out of memory
ex.observers.append(FileStorageObserver(train_exp_dir))
if torch.cuda.is_available():
device = torch.device('cuda')
torch.backends.cudnn.enabled = True
torch.backends.cudnn.benchmark = True
else:
device = torch.device('cpu')
@ex.automain
def my_main(_run, e, train_exp_dir, device, time_interval):
# load test file paths
test_list = list(np.load(train_exp_dir + '/test_list.npy'))
pred_exp_dir = train_exp_dir + '/%d'%(int(_run._id)) # prediction experiment directory
with open(train_exp_dir+'/config.json') as f:
config_train = json.load(f)
model = PhysNet(config_train['S'], config_train['in_ch']).to(device).eval()
model.load_state_dict(torch.load(train_exp_dir+'/epoch%d.pt'%(e), map_location=device)) # load weights to the model
# model.load_state_dict(torch.load("./demo/model_weights.pt", map_location=device))
@torch.no_grad()
def dl_model(imgs_clip):
# model inference
img_batch = imgs_clip
img_batch = img_batch.transpose((3,0,1,2))
# 在img_batch前面新增了一个批量大小的维度(批量大小为1)
img_batch = img_batch[np.newaxis].astype('float32')
img_batch = torch.tensor(img_batch).to(device)
rppg = model(img_batch)[:,-1, :] # (1, 5, T) -> (1, T)
rppg = rppg[0].detach().cpu().numpy()
return rppg
@torch.no_grad()
def read_wave(info, target_length):
bvp = []
hr = []
for signal in info["/FullPackage"]:
bvp.append(signal["Value"]["waveform"])
hr.append(signal["Value"]["pulseRate"])
T = len(bvp)
gts = list()
x_new = np.linspace(0, T - 1, num=target_length)
bvp_func = interpolate.CubicSpline(range(T), bvp)
gts.append(bvp_func(x_new))
hr_func = interpolate.CubicSpline(range(T), hr)
gts.append(hr_func(x_new))
return np.asarray(gts)
for h5_path in test_list:
h5_path = str(h5_path)
# print(h5_path)
with h5py.File(h5_path, 'r') as f:
imgs = f['imgs']
subject_name = os.path.basename(h5_path)[:-3]
json_path = f"/share2/data/PURE/{subject_name}/{subject_name}.json"
with open(json_path) as f:
info = json.load(f)
gts = read_wave(info, imgs.shape[0])
# print(gts.shape)
bvp = gts[0].reshape(-1, 1)
# bvppeak = f['bvp_peak']
fs = config_train['fs']
print(subject_name, imgs.shape, bvp.shape)
# duration表示秒数,fs表示frame per seccond
duration = np.min([imgs.shape[0], bvp.shape[0]]) / fs
num_blocks = int(duration // time_interval)
# 从整个视频当中截取出num_blocks个视频片段,这些片段之间是连续的(指从原视频当中截取的方式)
rppg_list = []
bvp_list = []
# bvppeak_list = []
for b in range(num_blocks):
rppg_clip = dl_model(imgs[b*time_interval*fs:(b+1)*time_interval*fs])
rppg_list.append(rppg_clip)
bvp_list.append(bvp[b*time_interval*fs:(b+1)*time_interval*fs])
# bvppeak_list.append(bvppeak[b*time_interval*fs:(b+1)*time_interval*fs])
rppg_list = np.array(rppg_list)
bvp_list = np.array(bvp_list)
bvp_list = bvp_list.reshape(num_blocks, -1)
print(num_blocks, rppg_list.shape, bvp_list.shape)
# bvppeak_list = np.array(bvppeak_list)
# results = {'rppg_list': rppg_list, 'bvp_list': bvp_list, 'bvppeak_list':bvppeak_list}
results = {'rppg_list': rppg_list, 'bvp_list': bvp_list}
np.save(pred_exp_dir+'/'+h5_path.split('/')[-1][:-3], results)