RuntimeError: shape '[2048, 49, 6]' is invalid for input of size 612500

[DB-123]

When I train, an error occurs.发生异常: RuntimeError
shape '[2048, 49, 6]' is invalid for input of size 612500
File "/media/DISK4TB/personal/Doubei/code_test/CDFSOD-benchmark-main/detectron2/modeling/meta_arch/devit.py", line 1287, in forward
indexes_change = indexes[cmask].view(bs, spatial_size, num_classes-1)
File "/media/DISK4TB/personal/Doubei/code_test/CDFSOD-benchmark-main/detectron2/engine/train_loop.py", line 273, in run_step
loss_dict = self.model(data)
File "/media/DISK4TB/personal/Doubei/code_test/CDFSOD-benchmark-main/detectron2/engine/defaults.py", line 510, in run_step
self._trainer.run_step()
File "/media/DISK4TB/personal/Doubei/code_test/CDFSOD-benchmark-main/detectron2/engine/train_loop.py", line 149, in train
self.run_step()
File "/media/DISK4TB/personal/Doubei/code_test/CDFSOD-benchmark-main/detectron2/engine/defaults.py", line 500, in train
super().train(self.start_iter, self.max_iter)
File "/media/DISK4TB/personal/Doubei/code_test/CDFSOD-benchmark-main/tools/train_net.py", line 202, in main
return trainer.train()
File "/media/DISK4TB/personal/Doubei/code_test/CDFSOD-benchmark-main/detectron2/engine/launch.py", line 82, in launch
main_func(*args)
File "/media/DISK4TB/personal/Doubei/code_test/CDFSOD-benchmark-main/tools/train_net.py", line 208, in
launch(
RuntimeError: shape '[2048, 49, 6]' is invalid for input of size 612500

[ohMargin]

check your input first, Are you currently reproducing the model or trying to fine tune it to your own dataset?
if your own dataset, check the num_classes, 612500 cannot be divided by 2048.
And please give more details about your error