Data Scientist - Traditional algorithm to classify the face features
Budget: ₹12,500 – ₹37,500 INR
Hi,
We are using mtcnn for detection and the Tensorflow facenet TFLite model for face recognition. The face recognition model returns 192 arrays. So the project architecture goes this way:
1. Face detection and crop images
2. Recognize face and get 192 array
3. Store this 192 array in database
4. While testing, detect face, get 192 array and find euclidean distance b/w test 192 array with all 192 array from database
5. Show results with least distance
WORK TO BE DONE:
1. Now rather than searching for all database embeddings, if we assign uniques id's or number or classifies database with some id numbers, if test images comes, the same way we will assign id for test images too and search only with a database of that id's
2. Assigning of these ids can be done with traditional or mathematical methods so that it can be implemented in java and swift
Note:
1. Right now we implemented multithreading, parallel processing, and database computation and achieved processing time under 3 seconds
2. Using the classification of arrays, we can achieve less than a second
Please apply if you can solve this challenge
We are using mtcnn for detection and the Tensorflow facenet TFLite model for face recognition. The face recognition model returns 192 arrays. So the project architecture goes this way:
1. Face detection and crop images
2. Recognize face and get 192 array
3. Store this 192 array in database
4. While testing, detect face, get 192 array and find euclidean distance b/w test 192 array with all 192 array from database
5. Show results with least distance
WORK TO BE DONE:
1. Now rather than searching for all database embeddings, if we assign uniques id's or number or classifies database with some id numbers, if test images comes, the same way we will assign id for test images too and search only with a database of that id's
2. Assigning of these ids can be done with traditional or mathematical methods so that it can be implemented in java and swift
Note:
1. Right now we implemented multithreading, parallel processing, and database computation and achieved processing time under 3 seconds
2. Using the classification of arrays, we can achieve less than a second
Please apply if you can solve this challenge