A serious bug in adet/modeling/solov2/solov2.py
DongChengdongHangZhou opened this issue · 2 comments
In 741 line, feature_add_all_level += self.convs_all_levelsi
Under certain pytorch and detectron environment, it will raise the Error as follows:
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [5, 128, 264, 256]], which is output 0 of ReluBackward0, is at version 3; expected version 0 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).
Becuase the inplace operation (+=) is not supported in some environment.
Please revise it as feature_add_all_level = feature_add_all_level + self.convs_all_levelsi
to increase the robustness of this code.
I also encountered this problem. Has this problem been solved
I also encountered this problem. Has this problem been solved
please follow my instruction as mentioned