Fangzhou Mu
Fangzhou Mu
It happens that the mini-ImageNet dataset (with input size 84x84) is small enough to fit in memory. Usually this means faster data loading but in the case of MAML, the...
It looks like either the LDI parameters or the test flags were not set correctly. I would suggest the following sanity checks one by one: - run ldi_render.py to see...
Thanks for the suggestion. The preprocessing code relies on multiple third-party tools. We will release it after some cleanup. Please check back in a few weeks.
Thanks for your interest. We did not put too much engineering effort on 3dphoto rendering, nor do we have access to the facebook code. We pre-processed input images based on...
@xingyi-li you are correct. We will update README this weekend and fix the broken links.
Thanks for your interest. Here is what the parameters mean in our provided mat files: - uv: pixel coordinates - z: per-pixel depth values from 3dphoto - rgb: per-pixel color...
The image center remains unchanged. We simply extend pixels beyond the borders to account for extrapolation. You can think of it as padding the input image with extrapolated pixels.
Please download the pre-trained models from this link first: http://pages.cs.wisc.edu/~fmu/gradfeat20/pretrained/ Then specify the correct file path in the config file. Let me know if there is still an issue.
Please make sure to set the --ndc flag. Also setting the -pc flag to 2 gives better results.
Thanks for the reminder. The links have been updated. The pre-processed COCO data is partitioned into parts. Please cat them before unzipping.