oichii / deeppulse-pytorch Goto Github PK

Hello, I'm planning on using the UBFC-rPPG dataset on physnet method, can you explain what is required for using such method? and is it possible to use a dataset that only has bpm as GT and video?

In pulse_data_3d.py, when printing the frame index ranges for each batch—determined by idx and idx+self.seq_len—there's an evident pattern. Initially, the iterations cover up to index 4010 before reverting to index 2151. As training progresses, this maximum index incrementally increases by several hundred. Essentially, the model retrains on the same frames. When using the whole PURE dataset, the training appears confined to the initial few videos, specifically 01-01, 01-02, and 01-03.

At line 44 of pulse_dataset_2d you initialise a reference csv.
reference = pd.read_csv(p + '.txt', sep='\t')
ref = reference.loc[:, 'waveform']
ref = np.array(ref)

How is structured that file? The numpy array does only contain the waveforms?
How can I extract heart rate during the test phase?

Thank you

Hi,
Would you please provide a sample of "train_seq.txt" ?!

Have you got any pretrained model? If I load the original spetlik hr-cnn models, will they work?

Thanks in advance

I keep getting this error while using the dataloader and realted scripts to load data. The sampler gives the error above, can you please guide me in what the problem is?

Some of the frames in the pure dataset do not have labelled data does the code cater for that?

May I have your pre-training model?Please! It is very helpful for what I am doing.