Restrictions

To convert a network model such as Faster R-CNN, YOLOv3, or YOLOv2 into an offline model compatible with the Ascend AI Processor, refer to Topics in the ATC Instructions to modify the .prototxt model file first.
Dynamic shape inputs are not supported, for example, NHWC = [?, ?, ?, 3] (for which you can specify any values to multiple dimensions). The dimension sizes must be static from model conversion and tuning.
Except constant operators, the quantity of input and output dimensions of operators at all layers in the model must not be 0.
Only the operators listed in the CANN Operator List are supported, and the operator restrictions must be met.
For details about the restrictions on model tuning, see "Tuning in Offline Inference Scenarios (Other Inference Devices) > Tuning Procedure" in the AOE Instructions.

Parent topic: Model Conversion and Tuning