Noah  Rowe

Noah Rowe

1595106480

Improve person re-identification with face detection (FaceBoxes)

Person re-identification is an interesting and not completely solved task. It includes finding (localizing) a person in an image and creating a digital description (vector or embedding) for a photo of a particular person in a way that the distance to the vectors for other photos of a particular person is closer than to the vectors generated for photos of other people.

Image for post

Person re-identification is used in many tasks including visitor flow analysis in a shopping center, tracking people across cameras, finding a certain person in a huge amount of photos.

Many effective models and approaches have been created recently to address the re-identification tasks. Full list of those models can be found here. But even the best models are still faced with a lot of problems, such as variations in pose and viewpoints of people because of which the embeddings for a photo of a person from different angles will be too far from each other, and the system can decide that this is a photo of different people.

The latest state-of-the-art models, such as Viewpoint-Aware Loss with Angular Regularization for Person Re-Identification, are designed to deal with mentioned problems, but we at ai-labs.org came up with a light approach that greatly simplifies the task of re-identification in some situations. I will talk about this approach in more detail.

Let’s start by explaining how most of the re-id frameworks detect photos of a particular person in the image. The most commonly used object detection models, such as Faster R-CNN or EfficientDet, are used to create a bounding box for the entire human body. After a photo for the entire human body is extracted, the embedding for this photo will be created.

Image for post

The problem is that object detection models often work even too well, they find photos of people from a variety of viewpoints and not always of the best quality. Embeddings based on these photos often do not allow correct re-identification of a person or are generated in such a way that embeddings for a photo of a particular person from one viewpoint will be close to embeddings for photos only from the same viewpoint, but not to embeddings for photos of the same person from a different viewpoint and distance.

#computer-vision #deep-learning #machine-learning #ai #deep learning

What is GEEK

Buddha Community

Improve person re-identification with face detection (FaceBoxes)

Top 6 Alternatives To Hugging Face

  • With Hugging Face raising $40 million funding, NLPs has the potential to provide us with a smarter world ahead.

In recent news, US-based NLP startup, Hugging Face  has raised a whopping $40 million in funding. The company is building a large open-source community to help the NLP ecosystem grow. Its transformers library is a python-based library that exposes an API for using a variety of well-known transformer architectures such as BERT, RoBERTa, GPT-2, and DistilBERT. Here is a list of the top alternatives to Hugging Face .

Watson Assistant

LUIS:

Lex

Dialogflow

#opinions #alternatives to hugging face #chatbot #hugging face #hugging face ai #hugging face chatbot #hugging face gpt-2 #hugging face nlp #hugging face transformer #ibm watson #nlp ai #nlp models #transformers

Noah  Rowe

Noah Rowe

1595106480

Improve person re-identification with face detection (FaceBoxes)

Person re-identification is an interesting and not completely solved task. It includes finding (localizing) a person in an image and creating a digital description (vector or embedding) for a photo of a particular person in a way that the distance to the vectors for other photos of a particular person is closer than to the vectors generated for photos of other people.

Image for post

Person re-identification is used in many tasks including visitor flow analysis in a shopping center, tracking people across cameras, finding a certain person in a huge amount of photos.

Many effective models and approaches have been created recently to address the re-identification tasks. Full list of those models can be found here. But even the best models are still faced with a lot of problems, such as variations in pose and viewpoints of people because of which the embeddings for a photo of a person from different angles will be too far from each other, and the system can decide that this is a photo of different people.

The latest state-of-the-art models, such as Viewpoint-Aware Loss with Angular Regularization for Person Re-Identification, are designed to deal with mentioned problems, but we at ai-labs.org came up with a light approach that greatly simplifies the task of re-identification in some situations. I will talk about this approach in more detail.

Let’s start by explaining how most of the re-id frameworks detect photos of a particular person in the image. The most commonly used object detection models, such as Faster R-CNN or EfficientDet, are used to create a bounding box for the entire human body. After a photo for the entire human body is extracted, the embedding for this photo will be created.

Image for post

The problem is that object detection models often work even too well, they find photos of people from a variety of viewpoints and not always of the best quality. Embeddings based on these photos often do not allow correct re-identification of a person or are generated in such a way that embeddings for a photo of a particular person from one viewpoint will be close to embeddings for photos only from the same viewpoint, but not to embeddings for photos of the same person from a different viewpoint and distance.

#computer-vision #deep-learning #machine-learning #ai #deep learning

Mia  Marquardt

Mia Marquardt

1619276400

FacePDFViewer— A PDF Viewer Controllable By Head Movements using Facial Landmark Detection

A real-time web tool using face-api.js to scroll through a PDF document in the browser using head movements.

Face Landmark Detection has many useful applications in Computer Vision such as face swapping, blink detection, face alignment, emotions recognition and head pose estimation.

In this article, I will show you my project “FacePDFViewer”. I have built a simple tool that allows you to move through a PDF document on the screen without using the mouse, just the movements of the head.

You can find the source code with instructions to run it in my GitHub repository, and a live demo is available here on GitHub.io.

The software is written in _Javascript _using face-api.js for face landmark detection and pdfjsfor the PDF documents rendering.

Project structure

The software consists of 3 real-time operations:

  1. Face landmark detection
  2. Head pose estimation
  3. PDF rendering

The PDF rendering is made with the open source _pdfjs _library that uses an html _canvas _element as a container of the document. In this article, I will talk about the first two operations, while I refer to pdfjs examples for PDF rendering, since my usage of the library in this project is very basic.

#face-api #tensorflow #computer-vision #face-detection-app #face-landmarks

Agnes  Sauer

Agnes Sauer

1595578440

Why we need person re-identification?

In this article, I will briefly introduce to you the field of tracking and re-identification without going into technical details. Before talking about re-identification it is essential to mention what person identification (tracking) is and how it works. Tracking is what the security people are responsible for with the only difference being that the machines do all the work. Therefore, a computer receives either some pre-recorded videos from surveillance cameras or real-time video and attempts to differentiate people and classify them. With the help of tracking, we can see the shapes of every person in the scene and identify their movements. This is all great but there are several issues with tracking in real-world scenarios…

Multi-person tracking issues

While tracking allows us to receive all the trajectories of movement of anyone in the scene and identify one person from another, the issues start to appear when we have multiple cameras. If the same person moves across a shopping mall and, for example, takes off his jacket in-between cameras, he will not be recognized. Different poses, outfits, backpacks, and other details can mess up our model and recognize the same person as two different ones.

Re-identification

Image for post

R_e-identification(reID)_ is the process of associating images or videos of the same person taken from different angles and cameras. The key to the issue is to find features that represent a person. Many of the recent models use deeply learned models to extract features and achieve good performance. Certain state-of-art methods are proposed based on convolutional neural networks (CNN), due to its powerful feature learning abilities and fitting capacity.

Good Practices for reID

Image for post

#deep-learning #tracking #reid #re-identification #computer-vision #deep learning

Nat  Kutch

Nat Kutch

1596723660

Exploring Other Face Detection Approaches

In these series of articles, we are exploring various other approaches of detecting faces rather than the common ones. In the previous article (Part1 and Part 2), we discussed about RetinaFace and SSH.

In this part we’ll discuss about _PCN: Progressi_ve Calibration Networks.

Image for post

PCN: Progressive Calibration Networks

Rotation invariant face detection i.e detecting rotated faces is widely required in unconstrained applications but still it remains a challenging task due to large variation of face appearances.

For addressing this problem, Progressive Calibration Networks(PCN) performs rotation-invariant face detection in a coarse-to-fine manner. PCN consists of three stages which detects faces and also calibrates RIP(Rotation in Plane) orientation of each face candidate to upright progressively.

PCN first calibrates those face candidates which as facing down to up, halving the range of RIP angles from [-180, 180] to [-90, 90]. Then the rotated faces further calibrated to upright range [-45, 45], halving RIP again. And then PCN makes the final decision for each face candidate and predicts precise RIP angle. It is discussed in detail below.

Architecture

Image for post

Proposed Framework

On an image all face candidates are obtained according to sliding window and image pyramid principle and each candidate window goes through the detector stage by stage [Here, 3 stages]. In each stage of PCN, the detector rejects faces with low confidence score, regresses bounding boxes of remaining face candidates and calibrates RIP. After each stage NMS is used to merge highly overlapping candidates.

PCN-1 1st Stage

For each input window x, PCN-1 has three objectives i.e. face/no face classification, bounding box regression and calibration.

The first objective which aims for classifying face and non-faces is achived by Softmax loss :

Image for post

where y equals to 1 if x if face, otherwise 0 and f is classification score.

The second objective attempts to regress the bounding boxes as :

Image for post

where t and t* represents the predicted and ground truth regression results and S is smooth L1 loss. The bounding box values consists of:

Image for post

where a, b and w denotes bounding box top-left coordinate and width. Variable without () are predicted values and with () are ground truth values.

#face-recognition #artificial-intelligence #face-detection-app #computer-vision #deep-learning #deep learning