haohang96

As illustrated in Figure1, _the leftmost query samples are from test classes, and the nearest neighbor samples are from training and test support set classes._ what's the meaning of **training...

I wondered why the step size $\epsilon$ of VP SDE in corrector need to multiply a more $\alpha_i$ compared with $\epsilon$ in VE SDE setting.

How to make sure $\hat z$ and $z$ in same location $s$ be positive pairs?

1

In Eq 3 of original paper, it seems that $\hat z^s$ and $z^s$ are positive pairs. It is reasonable in the 'horse' and 'zebra' examples because the location of object...

What's the meaning of "For fixedlarge, we set initial (log-)variance like os to get a better decoder log likelihood"?

1

Hi, I notice when you define the **model_variance** [here](https://github.com/openai/guided-diffusion/blob/22e0df8183507e13a7813f8d38d51b072ca1e67c/guided_diffusion/gaussian_diffusion.py#L281), first item of variance is set as **posterior_variance[1].** The comment here explains that _" we set initial (log-)variance like so to...

params.num_class in Baseline/Baseline++ method

3

In training stage of Baseline/Baseline++ method, it seems that you always set num_class as 200. But every dataset has specific number of base class(for example, miniImageNet has 64 base class)....

What's the meaning of "For fixedlarge, we set initial (log-)variance like os to get a better decoder log likelihood"?

1

Hi, I notice when you define the **model_variance** [here](https://github.com/openai/improved-diffusion/blob/783b6740edb79fdb7d063250db2c51cc9545dcd1/improved_diffusion/gaussian_diffusion.py#L279), first item of variance is set as **posterior_variance[1].** The comment here explains that _" we set initial (log-)variance like so to...

There is a numpy implementation of image perspective transformation: https://stackoverflow.com/questions/14177744/how-does-perspective-transformation-work-in-pil I wondered if the compose of four component: [rotation; translation; shear; projective] is equivalent to above perspective operation in numpy...

According to the official implementation of min-snr, in v-parametrization, the mse_loss_weight shall be (Ref to this line: [code-url](https://github.com/TiankaiHang/Min-SNR-Diffusion-Training/blob/baed0b2a9b18d878731c331033b5e76c184b76e4/guided_diffusion/gaussian_diffusion.py#L861C1-L864C104)): $$min(snr, \gamma) / (snr + 1)$$ But in your implementation, it is:...