Update for tflite import and relay import
* tflite import: handle squeeze operator as a reshape layer
* relay import: remove dependency on TVM
Signed-off-by: Hongmei Gou <a0271529@ti.com>
Signed-off-by: Jianzhong Xu <xuj@ti.com>
* tflite import: handle squeeze operator as a reshape layer
* relay import: remove dependency on TVM
Signed-off-by: Hongmei Gou <a0271529@ti.com>
Signed-off-by: Jianzhong Xu <xuj@ti.com>
TFLite: support subgraph partitioning and serialization of the partitioned model
* The subgraph supported by TIDL starts from the first operator of the model,
and by default stops before the first operator which is not supported by TIDL
* To overwrite the default, TIDL subgraph can be specified via passing
the output tensor name from the command line
* Only a single subgraph is currently supported
* During the serialization, the TIDL subgraph is represented as a TFLite custom
operator "tidl-am5-custom-op". The operators which are not part of the TIDL
subgraph are kept the same as in the input TFLite model
Signed-off-by: Hongmei Gou <a0271529@ti.com>
* The subgraph supported by TIDL starts from the first operator of the model,
and by default stops before the first operator which is not supported by TIDL
* To overwrite the default, TIDL subgraph can be specified via passing
the output tensor name from the command line
* Only a single subgraph is currently supported
* During the serialization, the TIDL subgraph is represented as a TFLite custom
operator "tidl-am5-custom-op". The operators which are not part of the TIDL
subgraph are kept the same as in the input TFLite model
Signed-off-by: Hongmei Gou <a0271529@ti.com>
TFLite update:
- fix problem of having two out data layers for tflite mobilenet models.
- write TIDL converted TFLite model back in .tflite format
Signed-off-by: Jianzhong Xu <xuj@ti.com>
- fix problem of having two out data layers for tflite mobilenet models.
- write TIDL converted TFLite model back in .tflite format
Signed-off-by: Jianzhong Xu <xuj@ti.com>
1. Add support for TensorFlow Lite (floating point models only).
2. Conv2d kernel types are automatically set when models are imported, and do not need to be explicitly configured.
2. Conv2d kernel types are automatically set when models are imported, and do not need to be explicitly configured.
Added support for mobilenetv2
1. fix eltwise layer import bug - PLSDK-2996
2. fix layer splitting issue for TensorFlow import - TIDL-527
3. update TIDL import config files to have optimal conv types - plsdk-2986
Signed-off-by: Jianzhong Xu <xuj@ti.com>
2. fix layer splitting issue for TensorFlow import - TIDL-527
3. update TIDL import config files to have optimal conv types - plsdk-2986
Signed-off-by: Jianzhong Xu <xuj@ti.com>
1. Added support for non-square padding for Caffe import.
2. Added check if Caffe model input layer format is supported.
Signed-off-by: Jianzhong Xu <xuj@ti.com>
2. Added check if Caffe model input layer format is supported.
Signed-off-by: Jianzhong Xu <xuj@ti.com>
1. TIDL import tool has been enhanced to capture and report all violations to TIDL limitations
2. Restriction on Inner Product Layer number of nodes has been removed
3. Protobuf has been updated to 3.6.1
Signed-off-by: Jianzhong Xu <xuj@ti.com>
2. Restriction on Inner Product Layer number of nodes has been removed
3. Protobuf has been updated to 3.6.1
Signed-off-by: Jianzhong Xu <xuj@ti.com>
TIDL import tool: 1. added support for ONNX.
2. added support for Tensorflow 1.7 and above.
3. added support for OpenCV for ARM build (not x86 build).
2. added support for Tensorflow 1.7 and above.
3. added support for OpenCV for ARM build (not x86 build).
Added protobuf interface files for ONNX support in ARMNN.
Signed-off-by: Jianzhong Xu <xuj@ti.com>
Signed-off-by: Jianzhong Xu <xuj@ti.com>
Added TIDL import tool source code.
Signed-off-by: Jianzhong Xu <xuj@ti.com>
Signed-off-by: Jianzhong Xu <xuj@ti.com>
caffeImport: Add Caffe LICENSE file for generated files in this folder only
Signed-off-by: Djordje Senicic <x0157990@ti.com>
Signed-off-by: Djordje Senicic <x0157990@ti.com>
Merge pull request #1 in PROCESSOR-SDK/tidl-utils from plsdk-2255 to master
* commit '09df882dc41aaf4abb7dc7c34dadad70ca62a483':
caffeImport: Add more details to readme installation instructions
caffeImport: Added interface files based on (arago) upstreamed version of protobuf 3.4.1
caffeImport: Add Caffe protobuf interface files, version 3.2
* commit '09df882dc41aaf4abb7dc7c34dadad70ca62a483':
caffeImport: Add more details to readme installation instructions
caffeImport: Added interface files based on (arago) upstreamed version of protobuf 3.4.1
caffeImport: Add Caffe protobuf interface files, version 3.2
caffeImport: Add more details to readme installation instructions
Signed-off-by: Djordje Senicic <x0157990@ti.com>
Signed-off-by: Djordje Senicic <x0157990@ti.com>
caffeImport: Added interface files based on (arago) upstreamed version of protobuf 3.4.1
Signed-off-by: Djordje Senicic <x0157990@ti.com>
Signed-off-by: Djordje Senicic <x0157990@ti.com>
caffeImport: Add Caffe protobuf interface files, version 3.2
Signed-off-by: Djordje Senicic <x0157990@ti.com>
Signed-off-by: Djordje Senicic <x0157990@ti.com>
Fix path in configuration files for target filesystem deployment
Signed-off-by: Djordje Senicic <x0157990@ti.com>
Signed-off-by: Djordje Senicic <x0157990@ti.com>
Initial commit of TIDL binary utilities: import tool and simulation tool, for Linux x86 and ARM Linux
Signed-off-by: Djordje Senicic <x0157990@ti.com>
Signed-off-by: Djordje Senicic <x0157990@ti.com>