TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Ask HN: Data augmentation tools for 3d object detection

28 pointsby iluvdataover 6 years ago
What are good data augmentation techniques which can work with 3d data specifically depth being third dimension. Also, any pointers on tools around them would be of great help.

3 comments

based2over 6 years ago
<a href="https:&#x2F;&#x2F;medium.com&#x2F;ymedialabs-innovation&#x2F;data-augmentation-techniques-in-cnn-using-tensorflow-371ae43d5be9" rel="nofollow">https:&#x2F;&#x2F;medium.com&#x2F;ymedialabs-innovation&#x2F;data-augmentation-t...</a><p><a href="https:&#x2F;&#x2F;www.quora.com&#x2F;What-are-good-data-augmentation-techniques-for-a-small-image-data-set" rel="nofollow">https:&#x2F;&#x2F;www.quora.com&#x2F;What-are-good-data-augmentation-techni...</a><p><a href="https:&#x2F;&#x2F;www.researchgate.net&#x2F;post&#x2F;Is_there_any_data_augmentation_technique_for_text_data_set" rel="nofollow">https:&#x2F;&#x2F;www.researchgate.net&#x2F;post&#x2F;Is_there_any_data_augmenta...</a><p><a href="https:&#x2F;&#x2F;forums.fast.ai&#x2F;t&#x2F;data-augmentation-for-nlp&#x2F;229&#x2F;15" rel="nofollow">https:&#x2F;&#x2F;forums.fast.ai&#x2F;t&#x2F;data-augmentation-for-nlp&#x2F;229&#x2F;15</a>
评论 #18627497 未加载
mathgaronover 6 years ago
Are you talking about a depth channel on an image plane? It also largely depend on the problem but here is a few tricks that helped me for my problems (object tracking).<p>- Generating synthetic data is powerful if you have the depth modality as it is easy to render. Also the real&#x2F;synthetic domain gap is narrow compared to RGB. I consider it as data augmentation: you usually do many renders from a single 3D model.<p>- If you can somehow normalize the offset (e.g. compute normals) that can help. In my case I could offset the center of the object as 0 depth and it greatly help the network to converge.<p>- Classic augmentations like gaussian noise, gaussian blur and also downsampling the depth helps (apply these randomly).<p>As for tooling, I just use numpy&#x2F;pytorch for most operations and OpenGL for renders.
lovelearningover 6 years ago
Regarding tools, OpenCV has a PLY-to-2D-images renderer [1].<p>[1]: <a href="https:&#x2F;&#x2F;github.com&#x2F;opencv&#x2F;opencv_contrib&#x2F;blob&#x2F;master&#x2F;modules&#x2F;cnn_3dobj&#x2F;samples&#x2F;sphereview_data.cpp#L83" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;opencv&#x2F;opencv_contrib&#x2F;blob&#x2F;master&#x2F;modules...</a>