ICCV 2019 Tutorial: Everything You Need to Know to Reproduce SOTA Deep Learning Models

Time: Sunday, October 27, 2019. Half Day - AM (0800-1215)
Location: Auditorium, COEX Convention Center

Presenter: Zhi Zhang, Sam Skalicky, Muhyun Kim, Jiyang Kang

AWS Icon   AmazonAI Icon   Neo Icon   Apache Incubator Icon   MXNet Icon   Gluon Icon   TVM Icon

Abstract

Deep Learning has become the de facto standard algorithm in computer vision. There are a surge amount of approaches being proposed every year for different tasks. Reproducing the complete system in every single detail can be problematic and time-consuming, especially for the beginners. Existing open-source implementations are typically not well-maintained and the code can be easily broken by the rapid updates of the deep learning frameworks. In this tutorial, we will walk through the technical details of the state-of-the-art (SOTA) algorithms in major computer vision tasks, and we also provide the code implementations and hands-on tutorials to reproduce the large-scale training in this tutorial.

Agenda

Time Title Slides Notebooks
8:00-8:15 Welcome and AWS Setup(Free instance available) link
8:15-8:40 Introduction to MXNet and GluonCV link,link
8:40-9:00 Deep Learning and Gluon Basics (NDArray, AutoGrad, Libraries) link,link
9:00-9:30 Bag of Tricks for Image Classification (ResNet, MobileNet, Inception) link link
9:30-10:00 Bag of Freebies for Object Detectors (SSD, Faster RCNN, YOLOV3) link link
10:00-10:30 Semantic segmentation algorithms (FCN, PSPNet, DeepLabV3, VPLR) link link
10:30-11:00 Pose Estimation(SimplePose, AlphaPose) link link
11:00-11:30 Action Recognition(TSN, I3D) link
11:30-12:00 Painless Deployment (C++, TVM) link link,link
12:00-12:15 Q&A and Closing

Q&A

Q1: How do I setup the environments for this tutorial?

A1: There will be all-in-one AWS SageMaker notebooks available for all local attendees, you need to bring your laptop and have a working email to access the notebooks.

Q2: How do I setup the environment in SageMaker after this tutorial?

A2: You can use lifetime-config to create sagemaker notebook instance using this lifetime-config. Make sure you have more than 30G disk space for the new notebook instance.

Organizers

Hang Zhang, Tong He, Zhi Zhang, Zhongyue Zhang, Haibin Lin, Aston Zhang, Mu Li