a2b3acfc6e
dnn: add the CANN backend ( #22634 )
...
* cann backend impl v1
* cann backend impl v2: use opencv parsers to build models for cann
* adjust fc according to the new transA and transB
* put cann net in cann backend node and reuse forwardLayer
* use fork() to create a child process and compile cann model
* remove legacy code
* remove debug code
* fall bcak to CPU backend if there is one layer not supoorted by CANN backend
* fix netInput forward
2022-12-21 09:04:41 +03:00
cdbb893b27
dnn: disable OpenCL code path in MatMul processing
...
- this mode is not supported by 22828
2022-12-20 09:46:48 +00:00
1102b7eff8
dnn: fix gather layer implementation
...
- support FP16 data
2022-12-20 06:09:34 +00:00
4891818114
make MatMul support 3D or 4D with broadcast
2022-12-15 10:36:08 +08:00
8ba44e7d55
Merge pull request #22882 from zihaomu:gemm_first_const
2022-12-08 14:18:33 +00:00
0a650b573b
Merge pull request #22840 from zihaomu:optimze_conv_memory_usage
...
DNN: reduce the memory used in convolution layer
* reduce the memory in winograd and disabel the test when usage memory is larger than 2gb.
* remove VERY_LOG tag
2022-12-08 12:57:13 +00:00
b16f76eede
Merge remote-tracking branch 'upstream/3.4' into merge-3.4
2022-12-03 12:39:41 +00:00
d16b3b2487
dnn(test): restore openvino tests with 'Cannot get memory' message
2022-12-03 01:34:48 +00:00
e14ca39fd7
Merge pull request #22857 from fengyuentau:batched_nms
...
dnn: add batched nms
2022-11-30 12:37:49 +03:00
421ba8730a
Merge pull request #22809 from fengyuentau:tile
...
dnn: support ONNX Tile
2022-11-29 14:42:28 +03:00
0d56524b72
gemm support transA and transB, and first input is constance.
2022-11-29 17:13:36 +08:00
9fded9ca53
batched nms impl
2022-11-29 15:32:34 +08:00
441624a5fb
tile impl
2022-11-29 11:15:38 +08:00
5044af69d1
let MatMul can work when both two inputs are const
2022-11-27 17:32:41 +08:00
ef2677b0a6
Make MatMul layer support 3d or 4d operation with const input
2022-11-10 11:41:44 +08:00
903bf0147e
Merge pull request #22666 from zihaomu:support_onnx_qdq_model
...
DNN: let Quant and Dequant of ONNX_importer support the Constant input.
* let Quant and Dequant support the Constant input.
* fix negative value of axis.
2022-10-31 16:06:31 +00:00
22f8fb4d5c
Do not fail tests in Yolo v7 model was not found.
2022-10-24 17:59:18 +03:00
35b2cff295
Merge pull request #22656 from dkurt:halide_fixes
...
* Fixes for Halide
* Enable some Halide tests
2022-10-21 17:49:49 +03:00
5d292826b2
Merge pull request #22593 from zihaomu:optimize_wino
...
optimize winograd futher more
2022-10-19 13:08:32 +03:00
f378f02954
Merge pull request #22652 from rogday:cuda_test_fixes
...
Address CUDA-related errors
2022-10-19 09:37:12 +03:00
dd14cf6a9c
address CUDA-related errors and enable cuda in elementwise ops
2022-10-18 16:54:42 +03:00
ec7fc5adca
Merge pull request #22529 from fengyuentau:scatter_scatternd
...
DNN: supports Scatter and ScatterND from ONNX
2022-10-17 14:57:46 +03:00
d24d8f2abe
implementation of scatter and scatternd with conformance tests enabled
2022-10-17 11:30:32 +08:00
d816442e4d
Make Unsqueeze layer support negative axes.
2022-10-14 18:00:19 +08:00
0fa43e3aac
Optimize the winograd futher more.
2022-10-14 10:15:45 +08:00
ec26541771
Merge pull request #22577 from zihaomu:Disable_winograd_branch_in_tryquantize
...
DNN: add enableWinograd API for Net
2022-10-11 09:44:00 +03:00
3419e64dcf
Merge pull request #22611 from zihaomu:greaterOrEqual
...
DNN: support GreaterOrEqual and LessOrEqual op in ONNX
2022-10-10 11:43:44 +03:00
1e2ceca4df
add enableWinograd API for Net.
2022-10-09 09:33:07 +08:00
347246901e
Merge pull request #21745 from alalek:dnn_plugin_openvino
2022-10-08 22:32:25 +00:00
9821fae59d
add greater_or_equal and less_or_equal ONNX support
2022-10-08 15:51:40 +08:00
43b2bb2c25
dnn: plugin support for OpenVINO
2022-10-07 16:57:31 +00:00
4557971481
enhance slice layer
...
refactor the code for parsing Slice layer
add test for Slice layer
let 'begin' and 'end' resize to dims
add opset message comment
2022-10-01 17:12:07 +08:00
a6274647a4
Merge pull request #21738 from rogday:gather
...
add Gather implementation
2022-09-19 16:21:14 +03:00
65f71ce2eb
add Gather implementation
2022-09-19 15:06:44 +03:00
6aefb8e86f
Merge pull request #22290 from fengyuentau:naive_yolov7
...
Support for YOLOv7 ONNX (not simplified)
2022-09-19 14:43:18 +03:00
4aef9b1c93
dnn: support yolov7 (not simplified)
2022-09-19 18:38:03 +08:00
337452b4c0
changed names of permutations if Reshpe is in NHWC
2022-09-03 19:02:41 +02:00
d2c48b898c
Merge pull request #22306 from zihaomu:qgemm_and_squeeze_opset13_onnximporter
2022-08-30 06:33:57 +00:00
2d837efba7
add qgemm and squeeze op13 supported on ONNXImporter
2022-08-30 09:50:29 +08:00
2619099fe5
Merge pull request #22337 from zihaomu:load_ONNX_fp16_as_fp32
...
DNN: load fp16 ONNX model as fp32
2022-08-29 09:32:25 +03:00
bb64db98d8
Further optimization of Conv2D, fused Conv_Add_Activation, bring latest code from ficus OpConv.fx. ( #22401 )
2022-08-26 12:57:25 +03:00
7eaec9dd22
load fp16 as fp32 and align fp16 and double in onnx_graph_simplifie
2022-08-26 10:04:44 +08:00
2ebdc04787
Merge remote-tracking branch 'upstream/3.4' into merge-3.4
2022-08-14 15:50:42 +00:00
bb71cb200e
Merge pull request #22199 from zihaomu:bug_fix_22195
...
DNN: Reduce Layer (add dynamic batch and ReduceSum support)
2022-08-11 12:59:51 +03:00
b2b7193374
Merge pull request #22311 from zihaomu:layer_fused_optmized_mish
...
DNN: add another two Mish activation to onnx_graph_simplifier
2022-08-05 14:22:06 +03:00
0614c40b42
add more skip for very long test case in test_dnn.
2022-08-02 14:58:05 +08:00
d4640f4647
support ReduceLayer without reshape layer.
2022-08-02 10:32:31 +08:00
3c5377ca1b
add another Mish graph simplifier.
2022-07-28 11:21:29 +08:00
ed69bcae2d
Merge pull request #21865 from rogday:nary_eltwise_layers
...
Reimplementation of Element-wise layers with broadcasting support
* init
* semi-working initial version
* add small_vector
* wip
* remove smallvec
* add nary function
* replace auto with Mat in lambda expr used in transform
* uncomment asserts
* autobuffer shape_buf & step_buf
* fix a missing bracket
* fixed a missing addLayer in parseElementWise
* solve one-dimensional broadcast
* remove pre_broadcast_transform for the case of two constants; fix missing constBlobsExtraInfo when addConstant is called
* one autobuffer for step & shape
* temporal fix for the missing original dimension information
* fix parseUnsqueeze when it gets a 1d tensor constant
* support sum/mean/min/max with only one input
* reuse old code to handle cases of two non-constant inputs
* add condition to handle div & mul of two non-constant inputs
* use || instead of or
* remove trainling spaces
* enlarge buf in binary_forward to contain other buffer
* use autobuffer in nary_forward
* generate data randomly and add more cases for perf
* add op and, or & xor
* update perf_dnn
* remove some comments
* remove legacy; add two ONNX conformance tests in filter
* move from cpu_denylist to all_denylist
* adjust parsing for inputs>=2
Co-authored-by: fengyuentau <yuantao.feng@opencv.org.cn >
2022-07-19 06:14:05 +03:00
1b8fba8e26
support ReduceSum with two input and dynamic shape batch size in ReduceLayer.
2022-07-13 13:46:16 +08:00
Yuantao Feng
Alexander Alekhin
zoom
Alexander Alekhin
Zihao Mu
Alexander Smorkalov
fengyuentau
Alexander Smorkalov
Dmitry Kurtaev
Smirnov Egor
anton