Implement a Pure Java AI-Based Palm Tracking Library
Budget: €30 – €250 EUR
I would like to develop a pure Java software library for AI-based palm tracking, designed to run entirely locally without relying on Python, native libraries (e.g., DLLs or shared objects), or external runtime dependencies. The final application should be distributed as a fat JAR, fully self-contained and ready to run on desktop systems.
Requirements:
1)The implementation must not use traditional techniques such as Haar Cascade Classifiers.
2)It is acceptable to use Java-based AI libraries, such as Deep Java Library (DJL){https://djl.ai/}, as long as they do not depend on native code.
3)The model should ideally be based on or trained using the EgoHands dataset{https://vision.soic.indiana.edu/projects/egohands/}, or compatible with it.
4)The library must detect palms in the input image and return their bounding box coordinates.
5)Detection of fingers or hand joints is considered a plus, but not strictly required.
6)The software is intended for desktop environments only (no mobile or embedded support required).
Output:
For each detected palm, the system should return the bounding box coordinates within the image.
Optionally, it may also return keypoint locations for finger joints if the model supports it.
Requirements:
1)The implementation must not use traditional techniques such as Haar Cascade Classifiers.
2)It is acceptable to use Java-based AI libraries, such as Deep Java Library (DJL){https://djl.ai/}, as long as they do not depend on native code.
3)The model should ideally be based on or trained using the EgoHands dataset{https://vision.soic.indiana.edu/projects/egohands/}, or compatible with it.
4)The library must detect palms in the input image and return their bounding box coordinates.
5)Detection of fingers or hand joints is considered a plus, but not strictly required.
6)The software is intended for desktop environments only (no mobile or embedded support required).
Output:
For each detected palm, the system should return the bounding box coordinates within the image.
Optionally, it may also return keypoint locations for finger joints if the model supports it.