티스토리 뷰
Udate 2021.08.24.
내가말한게 이거였음
https://github.com/pytorch/pytorch/issues/5059#issuecomment-404232359
np.random generates the same random numbers for each data batch · Issue #5059 · pytorch/pytorch
For example, the following script from __future__ import print_function import numpy as np import random # np.random.seed(0) class Transform(object): def __init__(self): pass def __call__(self, ite...
github.com
Reproducible PyTorch를 위한 randomness 올바르게 제어하기!
PyTorch 코드의 완벽한 재현을 위해 고려해야할 randomness를 제어하는 법을 설명드리겠습니다.
hoya012.github.io
아 근데 옛날에 실험할 때 numpy random seed를 고정해놓으면 thread마다 항상 같은 값을 내도록 됭어있어서 성능에 악영향이 있었던 것 같은데 확인해봐야겠다. 의도한 것은 thread마다도 다른 값을 내야됨. 정확히 문제가 뭐였는 지 기억이 안남. 그때 블로그에 적어둘걸..
괜찮은듯?
'Research (연구 관련)' 카테고리의 다른 글
scatter min (0) | 2021.08.24 |
---|---|
remote tensorboard (0) | 2021.08.14 |
DP vs DDP (0) | 2021.08.13 |
osmesa, pyrender, torch training bug (0) | 2021.08.13 |
ray & 3d bounding box intersection (0) | 2021.08.09 |
공지사항
최근에 올라온 글
최근에 달린 댓글
- Total
- Today
- Yesterday
링크
TAG
- VAE
- pyrender
- 헬스
- Interview
- spin
- 컴퓨터비전
- Transformation
- Virtual Camera
- demo
- 머신러닝
- 에디톨로지
- 피트니스
- focal length
- Machine Learning
- Generative model
- nohup
- Docker
- nerf
- 컴퓨터비젼
- 비전
- pytorch
- world coordinate
- camera coordinate
- 2d pose
- densepose
- part segmentation
- deep learning
- 문경식
- Pose2Mesh
- 인터뷰
일 | 월 | 화 | 수 | 목 | 금 | 토 |
---|---|---|---|---|---|---|
1 | 2 | 3 | 4 | 5 | ||
6 | 7 | 8 | 9 | 10 | 11 | 12 |
13 | 14 | 15 | 16 | 17 | 18 | 19 |
20 | 21 | 22 | 23 | 24 | 25 | 26 |
27 | 28 | 29 | 30 |
글 보관함