Hi,
I’m running the code from this repository https://github.com/junfu1115/DANet, Its a semantic segmentation network. The inference time taken was a lot so I was just looking into the code and found out that the part where the data is loaded to the model - Please refer this line, https://github.com/junfu1115/DANet/blob/201d308581d9d5a780a99f425eb5aee53bf3680b/encoding/models/base.py#L100 , takes 1.3secs. The only major operation performed here is shifting the data from cpu to gpu. So I was wondering why this is taking soo much time.
Also, the author has instructed to install pytorch from source using a commit-fd25a2a.
Also, I’m running with a single GPU - Tesla k80.
Can someone help me with this?