yhl1010

量化的IAO中用BN融合训练出现梯度爆炸

12

代码没有做任何改动，将bn_fold设为1训练会出现loss=nan的情况。修改了batch size 和学习率都一样，请问怎么解决呢？

How to generate sparse depth images and dense depth images from lidar point cloud?

3

sorry to bother you again, could you share with me the code How to generate sparse depth images and dense depth images from lidar point cloud? thank you very much!

about pre-trained model

1

I used the pre-training model you released and found that the evaluation indicators tested on the test dataset are better than the paper. Did you make some optimizations later or...

About Foreground Color Estimation

3

hello, thx for your great works. Do you have plans to release the training code or pre-trained models of Foreground Color Estimation?

How to specify which gpu to use for training

1

If I want to train only with the num'2' and num'3' gpus, how should i change the code? thank you very much!

when using colmap feature_extractor in command line, what the relationship between the arguments "ImageReader.default_focal_length_factor" and EXIF focal length of the pictures. Hope you can answer this question, thanks very much！

Is this a problem caused by the tensorflow version?

2

Initializing models: 0%| | 0/5 [00:00

Question about command line arguments

2

when using colmap feature_extractor in command line, what the relationship between the arguments "ImageReader.default_focal_length_factor" and EXIF focal length of the pictures. Hope you can answer this question, thanks very much！