Edgar Andrés Margffoy Tuay
Edgar Andrés Margffoy Tuay
Update: It seems that we got to replicate the old SRU results by using the new one without rescaling parameter
I've actually tried to implement my own dataset according to the VOC dataloader. However at training time, I got NaN values at loss
@alexiskattan I wrote a Visual Genome dataset loader: https://github.com/andfoy/textobjdetection/blob/master/visual_genome_loader.py#L482
@jtoy I'm actually on process of writing it, so is not published yet!
Also, when implementing https://github.com/cupy/cupy/pull/7985, I found that `unique` for n-dimensions could be optimized by means of a space filling curve, if somehow we managed to map n-dimensional points to a...
According to https://learn.microsoft.com/en-us/windows/console/createpseudoconsole, ConPTY requires at least Windows Server 2019. Since I haven't been able to implement the dynamic loading on winpty-rs (the backend library used by pywinpty), the only...
If you are willing to maintain the new fork, then I'll move it to the organization later today
These are the benchmark results for the current implementation of `find_peaks_cwt` as of [133cf70](https://github.com/cupy/cupy/pull/7875/commits/133cf701d013d89dde7c42a9ea642106983a8a34). As we discussed previously, the potential for parallel gains here are minimal, due to the high...
I've just reverted the implementation to the SciPy reference one, the tests are passing locally and the benchmark results are not so different from the previous ones: | Size |...
@takagi I'll check on my local Windows setup for the last time, else we'll just disable this one there