RuntimeError: expected scalar type Half but found Float: issues when I tried to run DCN on pytorch without C++

[AndywithCV]

Since I cannot compile C++ in my environment because of the imcompatibility of pakages, I then tried to use DCN with pytorch, but I encounter and error as below:

File "/root/data/andy/newyolov8/ultralytics/ultralytics/nn/modules/block.py", line 708, in dcnv3_core_pytorch sampling_input_ = F.grid_sample( File "/root/miniconda3/envs/newyolov8benchmark/lib/python3.8/site-packages/torch/nn/functional.py", line 4244, in grid_sample return torch.grid_sampler(input, grid, mode_enum, padding_mode_enum, align_corners) RuntimeError: expected scalar type Half but found Float

Please help me to take a look!!!

[GoblinCraftman]

and in the first part def _get_reference_points，there are still float32 remain. You need to change it into float16 or an error:meshgrid need the same type. And so on, in issue dcnv3 in modules also have two float need to change into float16. Thank you for your thought.