Added full FP16 computation branch on ARMv8 platform, 1.5x faster than FP32 #22275(FP16 Winograd is still pending).Further increased DNN speed on ARM and X86 by improving convolution, covering 1D and 3D cases, supporting convolution+element-wise op fusion. Fixes in nary element wise layer about broadcast:.#23491 Fixes for Segment Anything Model by Meta.#23613 Reduce Refactor for robustness and potential follow-up improvements.#23401 support ONNX Sub, PRelu, ConvTranspose.#23319 support ONNX Split, Slice, Clip (Relu6) and Conv with auto_pad.ONNX: #23047 Layer normalization, #23219 GELU and #23655 QLinearSoftmax.Improved layers => supported more models:.#23604 Enabled DNN module build without Protobuf dependency.#23161, #23409 TFLite models support, including int8 quantized models.Summer update for OpenCV 4.x has been released.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |