TI Deep Learning Library User Guide
TIDL Supported layers

This page lists layers/operators supported in current TIDL version. By default, TIDL releases include import tool supporting Caffe, Tensorflow, ONNX, tflite. Import tool will translate/import the training result file(prototxt+caffemodel, pb, onnx) to TIDL format.

TIDL Supported Layers Overview

  1. Convolution Layer
  2. Spatial Pooling Layer
    • Average and Max Pooling
  3. Global Pooling Layer
    • Average and Max Pooling
  4. ReLU Layer
  5. Element Wise Layer
    • Add, Product and Max
  6. Inner Product Layer
    • Fully Connected Layer
  7. Soft Max Layer
  8. Bias Layer
  9. Concatenate layer
  10. Scale Layer
  11. Batch Normalization layer
  12. Re-size Layer (For Bi-leaner/Nearest Neighbor Up-sample)
  13. RelU6 layer
  14. Detection output Layer (SSD - Post Processing As defined in caffe-Jacinto and TF Object detection API)
  15. Arg max layer
  16. PReLU (One Parameter per channel)
  17. Slice layer
  18. Crop layer
  19. Flatten layer
  20. ShuffleChannelLayer
  21. Faster RCNN ROI Pooling (As defined in TF Object detection API)

Validated Networks on Host/PC Emulation and EVM

  • 19 Classification networks across (Caffe, tensorflow and ONNX)
  • 8+ SSD based Object detection network trained on Caffe, ONNX, tensorflow Object Detection API
  • 6+ Pixel Wise Segmentation models trained on Caffe, ONNX, tensorflow
  • Please refer the Sample Pre-trained CNN Models for TIDL for full List

Note
In the current release, Only models/networks listed in the Sample Pre-trained CNN Models for TIDL are supported . Any Network which is not included in this list will NOT be supported
More networks (Including custom networks) will be supported in the future releases.

Core Layers/Operators Mapping & Notes

No TIDL Layer Type Caffe Layer Type Tensorflow Ops ONNX Ops tflite Ops Notes
1 TIDL_ConvolutionLayer Convolution
ConvolutionDepthwise 		Conv2D
DepthwiseConv2dNative 		Conv CONV_2D
DEPTHWISE_CONV_2D 		Regular & depth-wise conv will be imported as conv.
For TF and tflite DepthwiseConv2dNative, depth_multiplier shall be 1 in Number of input channels > 1.
ReLU & BN & Pooling will be merged into conv to get better performance. 1x1 conv will be converted to innerproduct.
Validated kernel size: 1x1, 3x3, 5x5, 7x7.
if stride == 4, only support kernel == 11x11.
if stride == 2, kernel should less than 7.
if kernelH*kernelW*input channel/groupNum+enableBias % 64 != 0, performance is not optimal.
2 TIDL_BatchNormLayer BatchNorm FusedBatchNorm BatchNormalization ReLU & Scale & Bias & PReLU will be merged & imported as BN.
16bit is not optimal for current version. All the channel-wise Broad cast operations are mapped to BN now.
3 TIDL_PoolingLayer Pooling MaxPooling
AvgPooling
Mean 		MaxPool
AveragePool
GlobalAveragePool 		MAX_POOL_2D
AVERAGE_POOL_2D
MEAN 		Validated pooling size: 1x1(MAX, stride 1x1/2x2), 2x2, 3x3.
4x4 pooling is not optimal.
4 TIDL_EltWiseLayer EltWise Add
Mul 		Add
Mul 		ADD
MUL 		Only support SUM/MAX/PRODUCT.
Only support 2 inputs.
5 TIDL_InnerProductLayer InnerProduct MatMul Gemm FULLY_CONNECTED Input shape must be 1x1x1xN. Please use global pooling/flatten before innerproduct.
Feature size larger than 2048*2048 is not optimal.
6 TIDL_SoftMaxLayer SoftMax Softmax Softmax SOFTMAX Input shape must be 1x1x1xN. Please use global pooling/flatten before softmax.
7 TIDL_Deconv2DLayer Deconvolution Conv2DTranspose ConvTranspose TRANSPOSE_CONV Recommend to use Resize/Upsample to get better performance. Only 4x4 kernel with 2x2 stride is supported
8 TIDL_ConcatLayer Concat ConcatV2 Concat CONCATENATION Concat will do channel-wise combination by default. Concat will be width-wise if coming after a flatten layer. used in the context of SSD
9 TIDL_SliceLayer Slice Slice Split NA Only support channel-wise slice.
10 TIDL_CropLayer Crop NA NA NA
11 TIDL_FlattenLayer Flatten NA Flatten NA 16bit is not optimal in current version.
12 TIDL_ArgMaxLayer ArgMax Argmax ArgMax ARG_MAX Only support axis == 1, mainly for the last layer of sematic segmentation.
13 TIDL_DetectionOutputLayer DetectionOutput tensorflow Object Detection API NA NA Please refer to comment 1.
14 TIDL_ShuffleChannelLayer ShuffleChannel NA Reshape + Transpose + Reshape NA
15 TIDL_ResizeLayer NA ResizeNearestNeighbor
ResizeBilinear 		UpSample RESIZE_NEAREST_NEIGHBOR
RESIZE_BILINEAR 		Only support 2x2, 4x4.

Other compatible layers

No Caffe Layer Type Tensorflow Ops ONNX Ops tflite Ops Notes
1 Bias BiasAdd Bias will be imported as BN.
2 Scale Scale will be imported as BN.
3 ReLU Relu
Relu6 		Relu RELU
RELU6 		ReLU will be imported as BN.
4 PReLU Prelu PReLU will be imported as BN.
5 Split Split Split layer will be removed after import.
6 Reshape Reshape Reshape RESHAPE Please refer to comment 1.
7 Permute Please refer to comment 1.
8 Priorbox Please refer to comment 1.
9 Pad Pad PAD Padding will be taken care of during import process, and this layer will be automatically removed by import tool.
10 MINIMUM For relu6 / relu8
11 DropOut This layer is only used in training, and this layer will be automatically removed during import process.
12 Squeeze Flatten
13 Shape Resize
14 Transpose For ShuffleChannelLayer
15 Clip Parametric activation threshold PACT
  • Comment 1: SSD(DetectionOutput/Permute/Priorbox/Reshape) usage
    • DetectionOuput, Permute, PriorBox, Reshape should be used along with each other. Standalone layer is not supported.
    • DetectionOutput layer is an important mark for SSD context. TIDL import tool will search for priorbox & permute & Reshape based on this layer.
    • Concat + Reshape + Softmax + DetectionOut in SSD context will be automatically change to Concat + DetectionOut. Reshape & Softmax will be integrated into DetectionOut layer automatically done by import tool.
    • PriorBox branch will be removed automatically and prior box info will be written into detection out layer. Only supported with Detection out layer in SSD context.
    • Permute Layer will be removed automatically by import tool. Permute operation will be done inside DetectionOutput Layer.
    • If DetectionOut layer is not used, Permute & PriorBox & Reshape are not allowed to use. You have to implement them all outside TIDL.
    • You can also use metaArch to implement SSD applications. Please refer to TIDL: metaArch User Guide .

For Unlisted Layers/Operators

Any unrecognized layers/operators will be converted to TIDL_UnsupportedLayer as a place-holder. The shape & parameters might not be correct. You may get the TIDL import result, but the result contains a unsupported layer, which will not work on target/PC. |

  • If this operation is supported by TIDL, but not supported by TIDL import tool:
    • Please modify the import tool source code.
  • If this operation is not supported by TIDL.