Multi-camera Data Labelling & Benchmark Creation
Budget: €250 – €750 EUR
I would start with a benchmark system for multi-camera data , focus on the re-id problem. You should look on: (1) standard for data labeling, (2) defined annotation structure of ground truth, (3) automation process for generating a benchmark.
Your task would be to develop a suitable semi-automated approach for labeling multi camera video content.
An idea could be that one is extracting with YOLO all person/cars (instance segmentation) of a video, collect all person/cars and propose a unique ID for similar person by using a single camera person tracker.
///////////////////////////////////////////////////////////////////////////////////////////
I am looking to achieve a set of comprehensive goals in multi-camera video content. The main tasks to be accomplished include -
1. Labeling of multi-camera/video data : Here we need an individual with skills in semi-automated (algorithm-assisted) labeling. You would be tasked with helping in the annotation of multi-camera/video data, a detailed understanding of the labeling standards is required.
2. Developing a semi-automated approach: The preferred methodology for labeling multi-camera/video content is through a semi-automated (algorithm-assisted) approach. We are interested in leveraging the YOLO (You Only Look Once) algorithm for instance segmentation. Therefore, a strong background with YOLO, and instance segmentation, would be the ideal candidate for the job.
3. Generating a benchmark for multi-camera data: This role's final and pivotal task includes creating an automated system for generating a benchmark for multi-camera data. Knowledge of the annotation structure of a ground truth is required for this aspect of the project.
At the end of this project, a standard should be established which can be used as a benchmark for multi-camera data metrics. This would be a valuable contribution to the re-id problem within the realm of multi-camera data. All interested candidates should have extensive experience with YOLO and a strong understanding of multi-camera/video content.
Your task would be to develop a suitable semi-automated approach for labeling multi camera video content.
An idea could be that one is extracting with YOLO all person/cars (instance segmentation) of a video, collect all person/cars and propose a unique ID for similar person by using a single camera person tracker.
///////////////////////////////////////////////////////////////////////////////////////////
I am looking to achieve a set of comprehensive goals in multi-camera video content. The main tasks to be accomplished include -
1. Labeling of multi-camera/video data : Here we need an individual with skills in semi-automated (algorithm-assisted) labeling. You would be tasked with helping in the annotation of multi-camera/video data, a detailed understanding of the labeling standards is required.
2. Developing a semi-automated approach: The preferred methodology for labeling multi-camera/video content is through a semi-automated (algorithm-assisted) approach. We are interested in leveraging the YOLO (You Only Look Once) algorithm for instance segmentation. Therefore, a strong background with YOLO, and instance segmentation, would be the ideal candidate for the job.
3. Generating a benchmark for multi-camera data: This role's final and pivotal task includes creating an automated system for generating a benchmark for multi-camera data. Knowledge of the annotation structure of a ground truth is required for this aspect of the project.
At the end of this project, a standard should be established which can be used as a benchmark for multi-camera data metrics. This would be a valuable contribution to the re-id problem within the realm of multi-camera data. All interested candidates should have extensive experience with YOLO and a strong understanding of multi-camera/video content.
Related categories:
Machine Learning (ML)
Data Science
Computer Vision
Machine Vision / Video Analytics
YOLO