Introduction to Object Detection How computers can see and how your Python app can, too. PyCon APAC

Introduction Who I am 2 Alan Descoins CTO @dekked_ | @tryolabs Introduction

3 Introduction Source: https://xkcd.com/1425/ (24 September 2014)

4 Introduction “In the 60s, Marvin Minsky assigned a couple of undergrads to spend the summer programming a computer to use a camera to identify objects in a scene. He figured they'd have the problem solved by the end of the summer. Half a century later, we're still working on it.”

Introduction Fast Forward to 2018: bird or not? # Initialize pre-trained model (ImageNet) model = ResNet50() # Load image and transform for model input x = load_img(img_path, target_size=(224, 224)) x = img_to_array(x) x = np.expand_dims(x, axis=0) x = preprocess_input(x) # Predict and decode preds = model.predict(x) print(Pred:', decode_predictions(preds, top=1)) 5 ● Python ● < 50 lines (with comments) ○ Core solver is just 7 lines ● Keras library ● ImageNet https://github.com/dekked/bird-or-not Bird! NOT Bird!

Agenda Introduction 6 Understanding images: the basics Intuitions behind modern object detection Luminoth: Python toolkit

Image classification Aka, label this picture for me, please.

What is so hard about this problem? Introducción 8 red green blue 1900 1300 Image classification

The Machine Learning approach 9 Image classification Dataset Algorithm (Convolutional NN) Trained model

Image classification Neural networks and Deep Learning Neural Networks for classification, widely used in the 80. Convolutional Neural Network (Yann LeCun, 1989) really good for pattern recognition with minimal preprocessing. 10 Handwritten digit recognition LeNet-5, Yann LeCun, 1998.

Image classification Convolutional filters A filter that looks at a small region and activates more strongly in the presence of certain pattern. 11 Several filters can detect more complex patterns:

Image classification The convolution operation Slide each filter through the image to produce an activation map. 12 Source: https://github.com/vdumoulin/conv_arithmetic Use more filters to detect patterns over activation maps (patterns over patterns over patterns…)

Image classification Remaining questions How do we know which filters/patterns to setup? → We learn them. They are regular weights of the network (use backpropagation). How do we know how many filters in each layer? → hyperparameter of the network (try and see what works best). 13 Source: https://cs231n.github.io/understanding-cnn/

Result of a convolutional layer 14 Image classification ... Original image 1900 x 1300 x 3 1900 x 1300 x 1 1900 x 1300 x 64

Summarizing information: pooling 15 Image classification ... ... 1900 x 1300 x 1 950 x 850 x 1 2x2 regions become 1x1 (max in each)

How (max) pooling looks like 16 Image classification Information corresponding to regions of the image is summarized.

Activation map Source: https://blog.heuritech.com/2016/02/29/a-brief-report-of-the-heuritech-deep-learning-meetup-5/ Visualizing a convolutional network Pre-trained: 17 Image classification

18 Finding interesting filters Image classification Learning combinations of filters that are activated (from the activation map) makes it a lot easier to find complex patterns!

Why didn’t this happen in the 80s? 19 Image classification ReLUs Dropout BatchNorm Skip-conn ...

From classification to detection Aka, tell me what and where, for all you see.

Object detection as supervised learning 21 Object detection person bicycle bird bird bird stick door (Pascal VOC)

Applications of object detection 22 Object detection CT scan of a lung cancer patient at the Jingdong Zhongmei private hospital in Yanjiao, China's Hebei Province (AP Photo/Andy Wong) Hsieh et al., “Drone-based Object Counting by Spatially Regularized Regional Proposal Networks”, ICCV 2017. Source: Pinterest

No content

Faster R-CNN Aka, Deep Learning model and its variants work really well.

Background Evolution of methods proposed in previous years: 25 Faster R-CNN 2014 R-CNN - Girshick et al. 2015 Fast R-CNN - Girshick. 2016 Faster R-CNN Ren, Girshick et al. “Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks”, CVPR 2016.

Faster R-CNN A two stage object detector 1. Propose interesting regions (RPN) Where should we look? 2. Analyze proposals & adjust (R-CNN) Is this an object? If so, which class? 26 person bicycle

First stage (1): Intuition behind regions 27 Faster R-CNN Combination of activation maps encodes spatial information!

First stage (2): Proposing regions 28 Faster R-CNN 1. Take a single spatial position. 2. Define reference box (anchor). 3. Learn if it’s an object or not, and adjust. 4. Repeat for every spatial position.

Anchor centers in original image Anchors reference (9 anchors per position) First stage (3): Visualizing anchor boxes All anchors 29 Faster R-CNN

First stage (4): resulting set of regions 30 Faster R-CNN Potentially hundreds or thousands!

31 Faster R-CNN Second stage (1): Using our proposals To learn how to adjust anchors, we looked at a small part of the activation map. To decide better, need to look at all activations corresponding to the regions.

32 Faster R-CNN Second stage (2): All proposals made equal Turn arbitrarily sized proposals into fixed size vectors / “squares”. Process is called RoI pooling. Allows us to feed into fully connected layer. 7 7

33 Faster R-CNN Second stage (3): Classification and adjustment Classification: what type of object is it (or is it background)? → probability distribution Regression: how should I resize the box to better enclose the object? → do it per class

Faster R-CNN Method summary ● Get an activation map with a CNN. ● Use it to propose interesting regions worth exploring. Associate an objectness score to them. ● Classify regions. Discard those that are background (ie. keep good scores only) Learn how to further adjust for each class of object. 34

Building a toolkit

What is Luminoth? Building a toolkit Open-source deep learning library/toolkit for computer vision object detection. 36 CLI tools Pre-defined models Cloud integration

Objectives Building a toolkit 37 “Out-of-the-box” usage Production ready Open source Readable code Extensible and modular

$ pip install luminoth $ lumi predict video.mp4 -k car Found 1 files to predict. Neither checkpoint not config specified, assuming `accurate`. Predicting video.mp4 [#############] 100% fps: 5.9 Simplicity as a goal 38 Building a toolkit

Data pipeline Debugging Training Data visualization Evaluation Deployment Beyond the model Distributed 39 Building a toolkit Unit testing Monitoring Model

Building a toolkit https://github.com/tryolabs/luminoth Using Luminoth 40 $ pip install luminoth $ lumi --help Usage: lumi [OPTIONS] COMMAND [ARGS]... Options: -h, --help Show this message and exit. Commands: checkpoint Groups of commands to manage checkpoints cloud Groups of commands to train models in the cloud dataset Groups of commands to manage datasets eval Evaluate trained (or training) models predict Obtain a model's predictions server Groups of commands to serve models train Train models

Invoking from your Python app 41 Building a toolkit from PIL import Image from luminoth.tools.checkpoint import get_checkpoint_config from luminoth.utils.predicting import PredictorNetwork image = Image.open('bird.jpg').convert('RGB') config = get_checkpoint_config('accurate') network = PredictorNetwork(config) objects = network.predict_image(image) # [{'bbox': [778, 323, 1124, 814], 'label': 'bird', 'prob': 0.9998}]

$ lumi dataset transform --type pascal --data-dir /data/pascal --output /data/ # Create tfrecords for optimizing data consumption. $ lumi train --config pascal-fasterrcnn.yml # Hours of training... $ tensorboard --logdir jobs/ # On another GPU/Machine/CPU. $ lumi eval --config pascal-fasterrcnn.yml # Checks for new checkpoints and writes logs. # Finally. $ lumi server web --config pascal-fasterrcnn.yml # Looks for checkpoint and loads it into a simple frontend/json API server. Luminoth use cycle 42 Building a toolkit

Luminoth: Demo

Learn more Learning more ● https://tryolabs.com/blog/ ● (Stanford) CS231n: Convolutional Neural Networks for Visual Recognition http://cs231n.stanford.edu/ ● Deep Learning (Goodfellow, Bengio, Courville) http://www.deeplearningbook.org/ ● https://distill.pub/ 44

Thanks for listening! Questions? github.com/tryolabs/luminoth @tryolabs @dekked_ Alan Descoins