learning phase

[xliucs]

Thanks for releasing the code for the meta-rPPG paper! Very interesting paper!

I am wondering where is the code for the learning phase described in the algrotihm 1 (Training of Meta-Learner)? Is it 'def fewshot_learning' whic is currently unifished? If so, does the trainng pipeline work without it?

Thanks!

[eugenelet]

Hi Xin,

Glad that our work is helpful. Thanks for taking interest in our work!

The adaptation phase and learning phase corresponds to L69 and L71 of train.py respectively. Comments are added in our latest commit.

Let us know if this answers your question.

[xliucs]

Hi Xin,

Glad that our work is helpful. Thanks for taking interest in our work!

The adaptation phase and learning phase corresponds to L69 and L71 of train.py respectively. Comments are added in our latest commit.

Let us know if this answers your question.

Hi Eugene,

Thanks for getting back to me.

According to the Algorithm 1 in the paper, the adaptation phase runs L steps whereas the learning phase run only once for each task. If L69 is adaptation phase, doesn't only run once?

[eugenelet]

Hi Xin,

You're mostly welcome!

The latest commit resolves this issue. We missed that line during the preparation of the code for public use. Thanks for pointing it out.

[xliucs]

Hi Xin,

You're mostly welcome!

The latest commit resolves this issue. We missed that line during the preparation of the code for public use. Thanks for pointing it out.

Thanks Eugene. That makes a lot more sense. BTW, it seems like you guys also ran the learning phase twice since you set opt.per_iter_task to be 3. I am curious why is it 3?

[eugenelet]

Hi Xin, that's just an ad-hoc decision used to sample 3 consecutive segments from a sequence of frames. We didn't tune opt.per_iter_task to fine-tune performance, further tuning might give better results.

[xliucs]

Thanks!

[eugenelet]

You're welcome!