My LinkedIn - / maltsevanton
My Twitter - / serious_wk
My Telegram channel - t.me/CVML_team
00:00:00 - Intro
00:00:29 - Official documentation
00:01:21 - Three approaches to inference: Single Core, Multi-core (inference vs switching)
00:02:15 - The common problem with speed measurement
00:03:37 - My Tests
00:04:47 - Numbers!
00:06:37 - Limitations of multi-thread approach
Three examples with multi-thread:
github.com/leafqycc/rknn-mult...
github.com/leafqycc/rknn-cpp-...
github.com/thanhtantran/rknn-...
Performance tweak - github.com/leafqycc/rknn-mult... (doesn't affect a lot for me just ~5%)
Rknn-pool manager - github.com/leafqycc/rknn-mult...
My inference sample - gist.github.com/ZlodeiBaal/dc...
Original one - github.com/leafqycc/rknn-mult...
Негізгі бет Ғылым және технология Maximizing Neural Networks Inference Speed on RockChip NPU
Пікірлер: 8