- 32
- 59 441
Shih-Shinh Huang
Приєднався 6 жов 2011
quarter DIP: Interactive Graph Cuts
This video introduces a graph-based interactive segmentation algorithm.
The outline of this lecture includes:
(1) Introduction about the motivation and problem description of interactive segmentation, and the central idea of the algorithm using graph cuts.
(2) Graph-based modeling about the construction of a graph, the cost definitions of the edges on the constructed graph, and the process of cost minimization for finding the optimal segmentation.
Any comments are welcome. (email: powwhuang@gmail.com)
All resources are available on the website (quarter.tw)
The outline of this lecture includes:
(1) Introduction about the motivation and problem description of interactive segmentation, and the central idea of the algorithm using graph cuts.
(2) Graph-based modeling about the construction of a graph, the cost definitions of the edges on the constructed graph, and the process of cost minimization for finding the optimal segmentation.
Any comments are welcome. (email: powwhuang@gmail.com)
All resources are available on the website (quarter.tw)
Переглядів: 35
Відео
quarter CNN: Temporal Convolution Networks (TCN)
Переглядів 62110 місяців тому
This video introduces the temporal convolutional network (TCN), a 1D convolutional neural network designed for sequence modeling tasks. The outline of this lecture includes: (1) Introduction about the motivation, central idea, and architecture overview of the TCN. (2) 1D dilated convolution about the definition and padding rule of the convolution process in TCN. (3) Residual block about the two...
quarter RNN: Simple Recurrent Neural Network (RNN)
Переглядів 168Рік тому
This video introduces the recurrent neural network (RNN) in its simplest form which has been widely used for dealing with sequential data. To clearly explain what the RNN is, we use the problem of price forecasting as an example to explain it in a step-by-step manner. The outline of this lecture includes: (1) Introduction about the RNN model, its main concept, and an example of price forecastin...
quarter RNN: Tokenization in Transformer From Text to Tokens
Переглядів 150Рік тому
This video talks about the tokenization used in Transformer to convert text into tokens for natural language processing. The outline of this lecture includes: (1) Introduction about what the Transformer is and what tokenization is. (2) Tokenization strategy about the main steps and the pros and cons of three tokenization algorithms, character, word, and subword. Any comments are welcome. (email...
quarter CNN: (YOLO v1) You Only Look Once Unified Real-Time Object Detection
Переглядів 7382 роки тому
This video talks about YOLO version 1 short for You Only Look Once. YOLO v1 is a unified real-time object detection algorithm based on convolutional neural networks and was published in CVPR, in 2016. The outline of this lecture includes: (1) Introduction about the background and the idea of YOLO v1. (2) Unified detection about the anchor mechanism used for object detection, network architectur...
quarter: DIP Hough Transform A Case Study of Line Detection
Переглядів 1332 роки тому
This video talks about Hough transform which is a well-known algorithm for detecting the shape of objects from their boundary points. For detailing Hough transform, we take the line detection as a case study. The outline of this lecture includes: (1) Introduction about the history, purpose, and idea of Hough transform. (2) Detection algorithm about the two steps of Hough transform, that are, sh...
quarter CNN Mask R-CNN
Переглядів 3202 роки тому
This video introduces the Mask R-CNN which is a convolution neural network for instance segmentation. The outline of this lecture includes: (1) Introduction about what instance segmentation is and how the Mask R-CNN works. (2) RoI align about the objective, formal statement, and align steps. (3) Mask branch about the design concept, network architecture and the mask loss defined for training. 0...
quarter DIP Introduction to Kalman Filter
Переглядів 2682 роки тому
This video introduces Kalman filter that is a popular algorithm for estimating some unknown variables given the measurements observed over time. Kalman filter was proposed by R. E. Kalman in 1960 and has been widely used in many applications. The outline of this video includes: (1) Introduction about what is Kalman filter and its main idea. (2) Gaussian distribution about the brief introduction...
quarter CNN: Feature Pyramid Networks for Object Detection
Переглядів 1,4 тис.3 роки тому
This video introduces feature pyramid networks that produce a feature pyramid for addressing the scaling problem in the object detection area. The outline of this lecture includes: (1) Introduction about the purpose of object detection, the solution for detecting objects in multiple scales, and the main idea of FPN. (2) Network architecture about the main components of FPN including bottom-up p...
quarter Py: Basic Digital Image Processing Using OpenCV on Google Colab
Переглядів 6943 роки тому
This video introduces basic digital image processing using Opencv on Google Colab which is a free cloud service for machine learning. The outline of this lecture includes: (1) Google Colaboartory about why uses Colab for writing machine learning algorithms, how to start Colab, and how to write the first Colab program. (2) Basic image processing about what is OpenCV and how to write the programs...
quarter CNN: Deep Residual Network
Переглядів 3753 роки тому
This video introduces the deep residual network published in the paper entitled “Deep Residual Learning for Image Classification”, by Kaiming He in CVPR 2016. The outline of this lecture includes: (1) Introduction about the depth evolution in the literature of neural networks and the issue to be addressed when the neural network uses more layers. (2) Deep residual network about the main idea of...
quarter DIP: Determining Optical Flow: Horn and Schunck Method
Переглядів 2 тис.3 роки тому
This video talks about an algorithm proposed by Horn and Schunck in 1981 for determining optical flow between two consecutive images. The outline of this lecture includes: (1) Introduction about the motion and the definition of the optical flow. (2) Brightness constraint about the Taylor expansion for formula derivation of brightness constraint. (3) Horn-Schunck method about the additional smoo...
quarter CNN: RoI Pooling and Align
Переглядів 3,4 тис.3 роки тому
This video introduces two well-known feature extraction operations, called RoI pooling and RoI align, that are widely used in two-stage object detection or object segmentation. The outline of this lecture includes: (1) Introduction about the background of RoI pooling and RoI align and what the RoI feature extraction is. (2) RoI pooling about the overview and pooling steps for RoI pooling (3) Ro...
quarter DIP Two Pass Connected Component Labeling for Binary Image
Переглядів 1,1 тис.4 роки тому
In this video, we introduce a two-pass algorithm for connected component labeling. The outline of this video includes: (1) Introduction about what is connected component labeling and the definitions of connectivity and connected component. (2) Two-pass labeling about the overview of connected component labeling algorithm and the operations performed in pass one and pass two, respectively. 00:00...
quarter Py Beginner's Guide on PyTorch from Linear Regression
Переглядів 2924 роки тому
This lecture introduces how to write a PyTorch program for machine learning. In order to focus on why PyTorch makes machine learning programming easier and more intuitive, we stick with a simple and familiar problem: a linear regression with a single feature. The outline of this lecture includes: (1) Introduction about the description of machine learning, the basic programming paradigm for mach...
quarter CNN: Region Proposal Network (RPN)
Переглядів 9 тис.4 роки тому
quarter CNN: Region Proposal Network (RPN)
quarter CNN: FaceNet: A Unified Embedding for Face Recognition and Clustering
Переглядів 6 тис.4 роки тому
quarter CNN: FaceNet: A Unified Embedding for Face Recognition and Clustering
quarter DIP: Real time Foreground Background Segmentation Using Codebook Model
Переглядів 7985 років тому
quarter DIP: Real time Foreground Background Segmentation Using Codebook Model
quarter DIP: Histogram Matching (Specification)
Переглядів 8635 років тому
quarter DIP: Histogram Matching (Specification)
quarter DIP Support Vector Machine: Two Separable Classes
Переглядів 2905 років тому
quarter DIP Support Vector Machine: Two Separable Classes
quarter DIP Otsu Algorithm Optimal Global Thresholding
Переглядів 3 тис.5 років тому
quarter DIP Otsu Algorithm Optimal Global Thresholding
quarter DIP LBP Local Binary Pattern
Переглядів 8 тис.5 років тому
quarter DIP LBP Local Binary Pattern
quarter DIP HOG Histogram of Oriented Gradients
Переглядів 3,1 тис.6 років тому
quarter DIP HOG Histogram of Oriented Gradients
quarter DIP Gaussian Mixture Models for Background Subtraction
Переглядів 9 тис.6 років тому
quarter DIP Gaussian Mixture Models for Background Subtraction
quarter DIP HLID Histogram of Local Intensity Difference
Переглядів 1216 років тому
quarter DIP HLID Histogram of Local Intensity Difference
quarter DIP Efficient Graph Based Image Segmentation
Переглядів 3,7 тис.6 років тому
quarter DIP Efficient Graph Based Image Segmentation
Thank you. You explained very well and easy to understand.
Thanks for your recognition. You are so welcome. ^^
Very useful, sir!
Thanks a lot ^^
How can we have access to the slides?
please send an email to quarterhuang2018@gmail.com. We will reply u with the slide in pdf format.
Now, you can get the slide from the website "quarter.tw"
最小平方法的概念我懂，但還是不太懂（u,v）怎麼來的。理論上每個點都應該要有一個初步的（u,v）後續才可以求平均值才對
一開始u,v都當零矩陣算，一直迭代下去就能求
謝謝你的協助回覆，^^
How to connect u...I am interested in doing research under a broad area name graph theory based image segmentation after watching your video on graph based segmentation
You can connect me via email "quarterhuang2018@gmail.com"
I search many paper regarding this topic, but I didn't get it.The way you explained the background subtraction using GMM is fabulous. Thank you so much for making this video.
Indeed, I am glad that this video is helpful to u.
Best Explanation Sir
Thanks for your recognition. You are so welcome
Thank you very much for the informative content sir
I am so glad that this video is helpful to you
Thanks from India😇
Your are so welcome ^^
Thank u very much <3
You are so welcome ^^
Incredible explanation, thank you so much Mr. Huang!
I am glad that you like this video and thanks for your recognition. ^^
Can I ask for the PowerPoint file of this article?
You can download the ppt in pdf format via the link "gg.gg/quarter"
Hi Professor Shih-Shinh Huang, thank you for this explanation. Clear and helpful.
Thanks a lot ^^
Excellent video and explanation, I think this is the clearest explanation on youtube. I have a quick question. Is it possible to swap out the ROI pooling for ROI align in a Faster RCNN network. I believe it is more accurate and i've seen this proposed in a few academic papers. I'm trying to build a custom object detector based on Faster-RCNN but with additional branches to detect additional image properties like occulusion, along with the normal BBOX and classification branch. Many thanks
Since RoI pooling and RoI align are both for normalizing feature maps for further predicting, it is feasible to substitute RoI pooling with RoI align within the Faster R-CNN network. However, this substitution comes at the expense of increased computational complexity.
Thank you very much for the great explanation.
You are so welcome. ^^. We are so glad to have your recognition.
Thanks, sir. You're a great teacher !
Thanks for your compliment. I’m really into it! ^^
老師講的真好，敲碗老師做yolov2~v7的教學影片
很謝謝你的肯定，也很高興你喜歡這個影片，關於 yolov2 ~ yolov7，我們也正考慮要錄製相關的影片，但據我們目前的理解，他們主要是基於 yolov1 進行不同層面的改良，這種類型的影片有點難做 ^^，還在傷腦筋中。
You explained it so clearly under 30 minutes. Hands down the best professor.
Thanks a lot for your recognition. ^^
感謝教授的教學 這個教學真的是核心中的核心
很謝謝你的肯定，也很開心你喜歡這個影片。
Thank you again. Question#1: Why do you upsample by a factor of 2 and what is the purpose of upsampling in a top down fashion please at 13:10? Question#2: Also may you please add a little what is aliasing effect you mentioned in 13:50? Do you mean because we up-sampled the feature maps by 2, the features become very close to each other, which is the aliasing effect if I am not wrong? Why the 3x3 Conv filter you applied solves aliasing effect please? I am new to this area?
Sorry for so late to response. Up-sampling by a factor of 2 in a top-down fashion is to make the feature map with the same dimension to the bottom one. The two features maps with the same dimension can perform element-by-element addition so as to propagate semantics from top to bottom.
How FPN produces feature pyramid please in 6:05?
It propagates semantics from high level down to low level.
May I know please what is dense scale sampling in simple terms at 4:27? Also, why we should use a feature map from the last convolution neural network for prediction please?
The feature map at the last conv. layer is the most semantic and is suitable for detection purposes.
@@quarter2018 Thank you, so what about dense scale sampling at 4:27, may I know its purpose please?
Thank you very much. May I know pleae why you said at 4:16 that The detection of a large object is from the large pyramidal level while the detection of a small object is from the lower pyramidal level?
Since the low pyramidal level has a high resolution, it is feasible to detect small objects. In other words, high pyramidal levels obtained from convolution operations generally have a lower resolution because of sub-sampling. This makes the feature of the small objects in higher layers in-visible.
@@quarter2018 Thank you.
Great explanation, Congratulations!
Thanks for your recognition.
Thank you for explaining simplistic and understandable way. Please can you do video on KNN Model for Background Subtraction
Thanks for your recognition. Do you have references for KNN model for background subtraction?
Thank you !
You are so welcome ^^
Thank you very much :)!
You are so welcome ^^
Gooooooooooooooooooood very gooood - Go on
Thanks a lot ^^
nice...
Thanks a lot! ^^
Thanks for the lecture. It is very concise and clear.
Thanks for your recognition. You are so welcome ^^
thank Prof!
You are so welcome
sir..could you plse share the ppt link to download.U have provided the best explanation of RCN.
You can download our slide in pdf format from our web-site via the link gg.gg/quarter
Great job thank you 🙏 ❤
Thanks for your recognition. ^^
thank you prof shih-shinh huang
You are so welcome
Thanks for this online lectures about ML. I can't wait to dive in .
We are so glad that you like our video
This is a very good explanation. Thank you very much!
You are so welcome ^^
Thank you.
You are so welcome
The accent of oral English of Taiwanese and Mainland Chinese do have a lot in common which makes your presentation much easier for me to understand, thanks, buddy!
I am glad that you enjoy this video ^^
Im not sure I understand the anchor labeling and anchor sampling step in training. When you first select a random image do you generate random anchors throughout the image at random locations? Why don't we use the ground truth bounding box in training instead of the highest IOU anchor box with the ground truth.
(1) The anchors are fixed at each point of the conv. feature map but not randomly generated. (2) During training, you have to compute the difference (called loss) between ground-truth bounding boxes and predicted ones. The selected anchors are considered as the predicted bounding boxes for computing the loss.
Is d in deep cnn mean the number of hidden layer that used?
$d$ is the number of neurons used in the output layer that makes your output is a $d$-dimensional feature vector.
How can i khow how many hidden layer used in facenet?
The number of hidden layers is dependent on your application.
@@quarter2018 can i talk with you on whatsapp or any way if you can. And thank you for your answer😍🌹🌺
@@waleedaiad3411 We will open a google meeting (meet.google.com/xya-vuys-vfo) for discussion at 10:00 PM ~ 10:30 PM (UTC+8) every Wednesday.
@@quarter2018 ok i will open with you Thanks for your answer🌺🌹
Great explanation! Incredibly clear! Thank you
Thanks a lot for your comments
Although your oral English is pretty poor, your content is fairly clear
First, thanks for your comment "your content is fairly clear". For your suggestion, what we currently do is to provide subtitle. In the future, we will spend more time on how to improve our oral English. Thanks for your comments.
thanks for such detailed explanation
You are so welcome
Very clear.
Thanks a lot
this video deserves more visits
Thanks for your recognition.
Nice Explanation. Very well articulated. Thank you
You are so welcome
Clear & Complete explanation, Thank you
Thanks for your recognition.
Best video I have ever seen about RPN!!!! I dont understand this subject for days now all clear thank you a lot!!!! Please keep going with making videos you are the best!!
Thanks for your recognition. We will keep going with this.
I heartily request you to make similar videos about Single Object Tracking and Multople Object Tracking. Don't worry if it is too long. I would rather spend 2 hours in a single video rather than wasting time and effort in looking for other non-sense videos.
Do you have any reference about single or multiple object tracking?