Alexander Alekhin
94e92cd6c0
dnn(ocl): skip int8 tests due to memory access issues
2021-10-06 21:27:18 +00:00
Jebastin Nadar
cce78cc5e2
Merge pull request #20535 from SamFC10:onnx-q
...
dnn : int8 quantized layers support in onnx importer
* added quantized layers support in onnx importer
* added more cases in eltwise node, some more checks
* added tests for quantized nodes
* relax thresholds for failed tests, address review comments
* refactoring based on review comments
* added support for unsupported cases and pre-quantized resnet50 test
* relax thresholds due to int8 resize layer
2021-10-04 18:07:38 +00:00
Zihao Mu
9085b933d8
Merge pull request #20702 from zihaomu:tf_expand_dim_layer
...
Add ExpandDims layer of tf_importer.cpp
* Add ExpandDims to tf_importer.
* add -1 expand test case.
* Support different dimensions of input.
* Compatible with 5-dimensional NDHWC data
* Code align
* support 3-dim input.
* 3-dim bug fixed.
* fixing error of code format.
2021-10-04 16:37:38 +00:00
SamFC10
87ebf2e50b
fix illegal memory access in int8 convolution
2021-10-03 15:16:01 +05:30
Alexander Alekhin
37c3f0d8a0
Merge remote-tracking branch 'upstream/3.4' into merge-3.4
2021-10-02 17:57:18 +00:00
Alexander Alekhin
f977d10a19
dnn(ocl): fix conv DWCONV workgroup
2021-10-01 18:52:07 +00:00
Alexander Alekhin
846317ef37
dnn(ocl): fix conv BASIC workgroup
2021-09-29 14:55:46 +00:00
Alexander Alekhin
24fcb7f813
Merge remote-tracking branch 'upstream/3.4' into merge-3.4
2021-09-25 17:50:00 +00:00
rogday
38b9ec7a18
Merge pull request #20682 from rogday:min
...
* Add Min layer to CPU, OpenCL, Halide, Inference Engine, NGraph and CUDA
* fix indentation
* add min to fusion and halide tests; fix doc
2021-09-22 15:17:37 +03:00
SamFC10
9c5d7716e2
fix for unsqueeze opset version 13
2021-09-17 17:40:57 +05:30
Alexander Alekhin
46fd26e366
Merge pull request #20699 from alalek:dnn_perf_update_convolution_tests
2021-09-16 17:11:32 +00:00
rogday
c410d7a97d
Merge pull request #20671 from rogday:yolov4x-mish
...
Add support for YOLOv4x-mish
* backport to 3.4 for supporting yolov4x-mish
* add YOLOv4x-mish test
* address review comments
Co-authored-by: Guo Xu <guoxu@1school.com.cn>
2021-09-14 17:49:49 +00:00
YashasSamaga
50462dcdc6
fix effrank assert to allow input effrank <= output effrank
2021-09-13 20:44:33 +05:30
Alexander Alekhin
6e66a9222a
dnn(onnx): fix format specifier
2021-09-11 22:26:52 +00:00
Alexander Alekhin
c3ac834526
Merge remote-tracking branch 'upstream/3.4' into merge-3.4
2021-09-11 21:27:26 +00:00
Zihao Mu
51b03b87e6
BiasAdd could load Const from second place.
2021-09-11 15:34:41 +00:00
Alexander Alekhin
1aacb9bb15
dnn(perf): update convolution tests
2021-09-10 13:11:02 +00:00
Alexander Alekhin
6ace801418
Merge pull request #20661 from alalek:dnn_ocl_fix_gemm_like_kernel
2021-09-10 11:58:52 +00:00
rogday
d31b93b513
Merge pull request #20674 from rogday:prelu_slope
...
Fix PReLU negative slope access pattern
* fix prelu negative slope access pattern
* change begin() to ptr()
2021-09-10 11:07:16 +00:00
rogday
4807cd8a6e
Merge pull request #20605 from rogday:split_slice_shenanigans
...
Add Normalize subgraph, fix Slice, Mul and Expand
* Add Normalize subgraph, support for starts<0 and axis<0 in Slice, Mul broadcasting in the middle and fix Expand's unsqueeze
* remove todos
* remove range-based for loop
* address review comments
* change >> to > > in template
* fix indexation
* fix expand that does nothing
2021-09-09 14:41:40 +03:00
Alexander Alekhin
35e824c287
dnn(ocl): fix out of bound access in GEMM-like kernels
...
- dropped usage of CreateSubBuffer() - buffers lifetime management issue
- fixed elementwise offset
- avoid out of bounds read access
2021-09-06 18:17:21 +00:00
Alexander Alekhin
5578ad5e14
dnn(ocl): fix automatic globalsize adjusting
...
- if kernel code doesn't support that
2021-09-06 03:11:29 +00:00
Alexander Alekhin
0a43b23275
Merge pull request #20651 from alalek:issue_18361
2021-09-04 18:22:12 +00:00
Alexander Alekhin
7967683296
Merge pull request #20648 from alalek:issue_20615
2021-09-04 18:21:58 +00:00
Alexander Alekhin
5b2c016834
dnn(ocl): avoid out of buffer access in copyWeightsSwizzled
2021-09-04 15:45:59 +00:00
Alexander Alekhin
407adc7061
dnn(ocl): fix buffer offsets in IDLF kernel
...
- drop CreateSubBuffer
- fix FUSED_CONV_ELTWISE mode
2021-09-04 15:28:35 +00:00
rogday
d0e612dc36
Merge pull request #20647 from rogday:resize_concat_optimization
...
Fix resize+concat optimization
* fix resize+concat optimization
* add comment and fix indentation
2021-09-03 12:32:29 +00:00
Alexander Alekhin
5aa7435d25
Merge remote-tracking branch 'upstream/3.4' into merge-3.4
2021-09-02 15:24:04 +00:00
Alexander Alekhin
060a76dc3e
Merge pull request #20573 from rogday:onnx_scale_fix
2021-09-01 14:09:17 +00:00
WJJ1995
edc442afdb
Merge pull request #20511 from wjj19950828:add_humanseg_support_0806
...
* support PPSeg model for dnn module
* fixed README for CI
* add test case
* fixed bug
* deal with comments
* rm dnn_model_runner
* update test case
* fixed bug for testcase
* update testcase
2021-09-01 10:10:05 +00:00
Alexander Alekhin
ae6fabc6fe
dnn(ocl): drop CL_KERNEL_PREFERRED_WORK_GROUP_SIZE_MULTIPLE check
...
- it is a hint and it should not block kernel execution
2021-08-30 20:40:14 +00:00
Alexander Alekhin
4c05a697fa
Merge remote-tracking branch 'upstream/3.4' into merge-3.4
2021-08-28 21:30:28 +00:00
Vincent Rabaud
38d0063c36
Do not use deprecated ReleaseCleared in protobuf library.
...
This is to make code work with protobuf arenas for memory
management (ReleaseCleared is incompatible).
The cleaning of the memory is also simpler.
2021-08-26 15:36:22 +02:00
Alexander Alekhin
6fbfc58602
Merge remote-tracking branch 'upstream/3.4' into merge-3.4
2021-08-21 17:25:18 +00:00
Alexander Alekhin
77a5c43d50
Merge pull request #20586 from alalek:issue_20585
2021-08-21 17:22:58 +00:00
Alexander Alekhin
f28e4b86fb
dnn(ocl): fix top initialization in verifyResult
2021-08-21 16:04:13 +00:00
rogday
6801dd043d
Merge pull request #20494 from rogday:onnx_diagnostic_fix
...
fix ONNXImporter diagnostic mode layer registration issue
* fix layer registration, thread unsafe access and align the behavior of DNN_DIAGNOSTICS_RUN between onnx and tf importers
* move skipModelInput
* print all missing layers
* address TF issue
2021-08-20 14:43:47 +00:00
Alexander Alekhin
a9817e9127
Merge pull request #20556 from rogday:onnx_split_sum_fix
2021-08-20 08:10:18 +00:00
Vincent Rabaud
9cfa84313c
Use the one argument version of SetTotalBytesLimit.
...
The two argument versions has been deprecated, cf
https://developers.google.com/protocol-buffers/docs/reference/cpp/google.protobuf.io.coded_stream
2021-08-19 14:31:29 +02:00
SamFC10
fa90e14b06
int8 layers and 8-bit quantization support
2021-08-19 09:56:47 +05:30
Smirnov Egor
fe625a558e
fix hasDynamicShapes for batch_size and fix axis selection in Scale layer
2021-08-18 19:22:24 +03:00
thezane
210bfaf8d6
Merge pull request #20483 from thezane:support-cumsum-layer-for-onnx
...
* Support cumsum layer for onnx
* Add unit tests
* Address review comments
2021-08-17 20:09:25 +03:00
Smirnov Egor
9ef41f68fb
fix Split partial sum
2021-08-16 15:44:54 +03:00
Alexander Alekhin
05d733e707
Merge pull request #20524 from yichenj:dnn_text_recognition_enhance
2021-08-15 12:30:25 +00:00
Alexander Alekhin
0c01cf7c85
Merge remote-tracking branch 'upstream/3.4' into merge-3.4
2021-08-14 18:24:00 +00:00
Julia Bareeva
cfb36443fb
Merge pull request #20506 from JulieBar:lstm_activations
...
* Support activations(Sigmoid, Tanh) for LSTM
* fix warning
2021-08-13 15:41:00 +03:00
Alexander Alekhin
9d3826c676
Merge pull request #20525 from SamFC10:fix-prior-variances
2021-08-13 10:06:55 +00:00
JIANG Yichen
955cf35d5f
Implement ctc prefix beam search decode for TextRecognitionModel.
...
The algorithm is based on Hannun's paper: First-Pass Large Vocabulary
Continuous Speech Recognition using Bi-Directional Recurrent DNNs
2021-08-12 20:33:31 +08:00
HAN Liutong
aaca4987c9
Merge pull request #20287 from hanliutong:dev-rvv-0.10
...
Optimization of DNN using native RISC-V vector intrinsics.
* Use RVV to optimize fastGEMM (FP32) in DNN.
* Use RVV to optimize fastGEMM1T in DNN.
* Use RVV to optimize fastConv in DNN.
* Use RVV to optimize fastDepthwiseConv in DNN.
* Vectorize tails using vl.
* Use "vl" instead of scalar to handle small block in fastConv.
* Fix memory access out of bound in "fastGEMM1T".
* Remove setvl.
* Remove useless initialization.
* Use loop unrolling to handle tail part instead of switch.
2021-08-11 01:16:03 +03:00
Smirnov Egor
739ff84732
add Max layer to TFImporter
2021-08-09 14:01:51 +03:00
