Post

Replies

Boosts

Views

Activity

Reply to Crash inside of Vision framework during VNImageRequestHandler use
Thanks for the answers. Currently the number of crashes is low, with most coming from 18.0.1, and none from 18.3 so far, so fixing this on-hold in the app I work on. I did a check of Vision/CoreML - and it seems quite straightforward, and removes one dependency layer. I think I'll try it even if the bug doesn't reoccur. I was considering running the models synchronously - as the newer devices get faster, there's not much gain from running models in parallel, but the bug occurrences slowed down. Since MLModel main documentation page states that it needs to be run on a single thread, I'd probably still do it - but then it allows batching, and it may solve the performance problem that the original code tried to solve by concurrency.
Topic: Machine Learning & AI SubTopic: General Tags:
Mar ’25