search

DerryHub / BEVFormer_tensorrt

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

Question about the fp16 mode of mmdeploy grid_sampler #94

Open kamiLight opened 10 months ago

kamiLight commented 10 months ago

It seems to mmdeploy grid_sampler do not support fp16 (i.e nvinfer1::DataType::kHALF) according to https://github.com/open-mmlab/mmdeploy/blob/bc75c9d6c8940aa03d0e1e5b5962bd930478ba77/csrc/mmdeploy/backend_ops/tensorrt/grid_sampler/trt_grid_sampler.cpp#L110, so i wonder that mmdeploy grid_sampler will do nothing just return 1 if using fp16 mode? In this case, wouldn't we get the wrong result? Thanks!

admyxs commented 8 months ago

It seems to mmdeploy grid_sampler do not support fp16 (i.e nvinfer1::DataType::kHALF) according to https://github.com/open-mmlab/mmdeploy/blob/bc75c9d6c8940aa03d0e1e5b5962bd930478ba77/csrc/mmdeploy/backend_ops/tensorrt/grid_sampler/trt_grid_sampler.cpp#L110, so i wonder that mmdeploy grid_sampler will do nothing just return 1 if using fp16 mode? In this case, wouldn't we get the wrong result? Thanks!

about this question

Do you know the answer?