TI Deep Learning Library User Guide
Supported Networks
  • This release is verified on multiple classification, object detection and segemntation networks. Please refer data sheet for more details

Supported Exchange Formats

  • This release is verified for CNN networks in below exchange formats
    • TIDL Network binary format
    • Caffe
    • TensorFlow
    • ONNX
    • TFLite

The network models are not provided with the package and can be downloaded from web. The current configuration files supplied as part of release package assumes models directory to be available at directory (../../test/testvecs/models/) relative executable directory. But user can have any directory structure to hold the models and modify the config files to point to appropriate location. The models which are verified can be downloaded from web and web links of these are provided at Model Details

For Tensor flow frozen graph in form of binary protobuf (*.pb) is expected. Tensor flow provides utilities (https://github.com/tensorflow/tensorflow/tree/master/tensorflow/python/tools) to generate this from a checkpoint. The details of how to generate frozen graph is out of scope for this document.

Supported Layers

Usage of other network model may result into undefined behavior