Reading list

Labels:
C : Common
A: ACENet only
M: Matting only

WEEK 1

1. Basics (C)
a. Go through the entire course (multiple times if possible) :
https://cs231n.github.io/
NOTE: If you think the volume is too much, start at "Neural Networks Part 1:
Setting up the Architecture"

2. Practice (C)
a. 60 minute tutorial :
https://pytorch.org/tutorials/beginner/deep_learning_60min_blitz.html

- Discussion session

WEEK 2

3. Architecture (C)
a. Survey : https://towardsdatascience.com/from-lenet-to-efficientnet-the-evolution-
of-cnns-3a57eb34672f
b. ResNet
i. Paper: https://arxiv.org/abs/1512.03385
ii. Blog: https://towardsdatascience.com/an-overview-of-resnet-and-its-
variants-5281e2f56035
c. MobileNetV2:
i. Paper: https://arxiv.org/abs/1801.04381
ii. Blog: https://towardsdatascience.com/review-mobilenetv2-light-
weight-model-image-classification-8febb490e61c
d. Decoder architecture :
i. Paper: https://arxiv.org/abs/1703.02719
ii. Blog: https://towardsdatascience.com/review-gcn-global-
convolutional-network-large-kernel-matters-semantic-segmentation-
c830073492d2

4. Loss Functions (C)

a. Loss functions : https://arxiv.org/abs/2006.14822
5. Metrics (C)
a. Confusion Matrix: https://towardsdatascience.com/understanding-confusion-
matrix-a9ad42dcfd62
b. ROC Curve: https://machinelearningmastery.com/roc-curves-and-precision-
recall-curves-for-classification-in-python/
c. IoU : https://towardsdatascience.com/intersection-over-union-iou-
calculation-for-evaluating-an-image-segmentation-model-8b22e2e84686
d. mAP : https://towardsdatascience.com/map-mean-average-precision-might-
confuse-you-5956f1bfa9e2

6. Practice (C)
a. MNIST Tutorial :
https://nextjournal.com/gkoehler/pytorch-mnist
b. MNIST > 99% test Accuracy

- discussion session

WEEK 3
7. Segmentation (A)
a. Basics:
i. What is Semantic Segmentation?
ii. What is Instance Segmentation?
iii. What is Panoptic Segmentation?
iv. What are the commonly used datasets for segmentation tasks?
v. Benchmark : https://paperswithcode.com/task/semantic-segmentation

b. Instance Segmentation
i. Mask-RCNN: https://arxiv.org/abs/1703.06870
ii. Mask-RCNN blog : https://towardsdatascience.com/computer-vision-
instance-segmentation-with-mask-r-cnn-7983502fcad1
iii. ACENet : https://arxiv.org/abs/1906.11109
iv. SOLO (optional): https://arxiv.org/abs/1912.04488
v. BANet (optional) : https://arxiv.org/abs/1901.03814

8. Matting (M)
a. Basics (Keep these concepts in mind while reading papers):
i. What is a Trimap?
ii. How do people usually create a Trimap?
iii. What is an Alpha Map and how it is different from a segmentation
map?
iv. What are the losses mostly used in benchmark?(MSE, SAD, Grad,
Connectivity) and why are they important? (Don't get into loss
implementation but what it conveys to matting task)

b. This blog post will help a bit to understand problem. But don't get into what he
explained about FBA Matting
https://learnopencv.com/image-matting-with-state-of-the-art-method-f-b-
alpha-matting/

c. Research papers (in order of reading):

 i.
Good beginning of DNN in matting: https://arxiv.org/abs/1703.03872
ii.
Base research paper for our work: https://arxiv.org/abs/1909.04686
iii.
Best research: https://arxiv.org/abs/2003.07711
iv.A research direction (to boggle the thought process) :
https://github.com/ZHKKKe/MODNet
v. Boiler plate code : https://github.com/Yaoyi-Li/GCA-Matting

d. Online benchmark: http://alphamatting.com/index.html

Tools

 Basic Tools:
o PyTorch: https://pytorch.org/get-started/locally/
o VSCode for Python : https://code.visualstudio.com/docs/languages/python
o MobaXTerm: https://mobaxterm.mobatek.net/download.html
o WinSCP: https://winscp.net/eng/download.php

 Advanced Tools:
o Augmentations: https://albumentations.ai/
o Hyper-parameter optimization: https://nni.readthedocs.io/en/stable/