search

DerryHub / BEVFormer_tensorrt

BEVFormer inference on TensorRT, including INT8 Quantization and Custom TensorRT Plugins (float/half/half2/int8).
Apache License 2.0
430 stars 71 forks source link

Reduce precision error of MSDA in INT8 mode #25

Closed DerryHub closed 1 year ago