Dan Rossi

I've confirmed there is no documented way to convert from the openmodels pytorch file. pnnx is broken. It doesn't document how to load the pth model either. https://github.com/pnnx/pnnx/issues/227#issuecomment-3135028889

I don't see docs there how to load pytorch for conversion to ncnn. But they do use spandrel which broke giving it to pnxx. This might detail how to load...

I had another look. there is no docs how to use their converter tool. Their spandrel library is broken immediately with any openmodels. And they document to load pytorch models...

I had a look at every documentation out there using pytorch to export to onnx first. They all skip a test. they are oddly trying to import the model as...

I just figured out not all models from openmodels is going to convert even as onnx but it almost half converted with pnnx. It's expecting a specific undocumented input shape....

That is an epic job converting the shader to nodes system. I tried myself but it was too difficult. However for webgpu to work `outputEncoding` needs to be changed to...

I was going to report on this. Wowza has just turned on 708 caption injection but there isn't a consolidated way to get players to support it. I reported to...

It seems to be an oversight. There is an event called before appending with not really useful information on the segment without the merged bytes with the init segment bytes....

It works for a while but then the init segment isn't in the data being appended. So needs an event to get the seperated init segment.

This patching is required to get the initsegment to merge back in ``` const playlistController = player.tech().vhs.playlistController_, //oldAppendBuffer = playlistController.sourceUpdater_.appendBuffer.bind(playlistController.sourceUpdater_), oldAppendData = playlistController.mainSegmentLoader_.appendData_.bind(playlistController.mainSegmentLoader_); playlistController.mainSegmentLoader_.appendData_ = (segmentInfo, result) => { oldAppendData(segmentInfo,...