Introduction to GluonCV
Download as PPTX, PDF1 like333 views
GluonCV is a framework designed to streamline deep learning processes by providing pre-trained models, training scripts, and easy-to-use APIs, aiming to facilitate the reproduction of results from important papers. It addresses common challenges faced by researchers, such as debugging complex models and adapting open-source implementations to specific tasks. GluonCV includes various tools for image classification, semantic segmentation, object detection, instance segmentation, and more, while also fostering community support through forums and GitHub.
Introduction to GluonCV
- 1. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark Introduction to GluonCV
- 2. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark Why GluonCV? • What is the biggest challenge you have ever encountered with deep learning?
- 3. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark Why GluonCV? • What is the biggest challenge you have ever encountered with deep learning? • “reproducing the best claimed results from latest papers” SOTA state-of-the-art
- 4. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark Real-world Stories #1 • Back to a period in 2016, the same ImageNet models trained by MXNet achieved on average 1% worse accuracy compared to Torch. • Tried almost everything to debug, even developed a plugin to run Torch code inside MXNet to make it easier to compare results. => Transcoding training images using 95 JPEG quality rather than 85 solved the problem.
- 5. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark Real-world Stories #2 • Using another open source DL framework: trained model accuracies cannot match previous internal version. • Spent months to figure out why, with no clue. => The order of data augmentation is different from previous version.
- 6. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark • I will write clean and reusable code when I’m prototyping this time. • Variant: • - I will write clean and reusable code next time. Common myth 1
- 7. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark Common myth 2 • My code will still run next year. • Sometimes, it’s not our fault.
- 8. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark Common myth3 • I will finish setting up the baseline model this afternoon. • Though it may not be our fault again.
- 9. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark Starting from scratch can be hard • Even the most talented researchers will get blocked by trivial things. • Experience and instincts can be your enemies in certain circumstances. • Training is time-consuming, initialization and augmentation is randomized, and many implementation details need to be taken care of. => Debugging deep learning models is extremely difficult.
- 10. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark
- 11. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark It’s not easy to embrace open-source implementations • Often the quality of open-source implementations vary. • Languages, code styles, project structures, DL frameworks are mixed. • Personal projects tend to focusing on a specific task with specific datasets. It requires significant engineering efforts to adapt to your use case. • Community projects can be abandoned frequently.
- 12. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark What does GluonCV provide • Reproduction of important papers in recent years • Model zoo with 80+ pre-trained models • Training scripts (as well as tuned hyper-parameters) to reproduce the results • Full training script + Dataset download script • Logs of training run
- 13. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark What does GluonCV provide • Considerate APIs and modules that are easy to follow and understand • Avoid re-writing the same utilities again and again • Pre-set data augmentation and transforms, visualization and training utilities • Community support, feel free to ask and discuss • User forum • Github community and open roadmap
- 14. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark
- 15. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark Image Classification • More than 50+ pre-trained ImageNet models(ResNet, MobileNet…) • We achieved the best accuracy using some of the most popular models (e.g., ResNet), compared with other frameworks • Used as backbone in many downstream tasks => better accuracy
- 16. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark Semantic Segmentation • FCN • PSPNet • Mask-RCNN • DeepLab
- 17. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark Object Detection • SSD and YOLOv3: fastest solution • Faster-RCNN, RFCN and FPN: slower but more accurate, especially for tiny objects
- 18. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark
- 19. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark Instance Segmentation • Mask R-CNN
- 20. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark Key Point Estimation • SimplePose
- 21. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark Style Transfer MSGNet GANs CycleGAN SRGAN WGAN Re-identification Market1501
- 22. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark Coming Soon: Depth Estimation
- 23. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Trademark Like GluonCV? Go build! https://gluon-cv.mxnet.io https://github.com/dmlc/gluon-cv
Editor's Notes
- #2: First call deck for a high level introduction to Apache MXNet.