Disc Golf Disc Recognition Model
Budget: $750 – $1,500 SGD
I need a computer-vision model that can look at a photo of a disc golf disc and instantly tell me which brand it is and, whenever the stamp allows, read the disc’s name as well. Beyond straight identification, the model has to understand subtle differences in colour schemes and patterns so that two visually similar discs from the same mould but in different plastics still register as distinct items.
During training the system must learn to:
• recognise the disc itself in any orientation, even when it fills only part of the frame;
• focus on brand logos first, using them as the primary classification key;
• capture colour and pattern data so the output can include a short textual description of what it sees;
• flag the general condition (new, lightly used, scratched, etc.) based on surface marks;
• parse the stamped text to return the disc name whenever it is legible.
A working solution should come with:
1. the annotated training/validation pipeline (Python preferred; TensorFlow or PyTorch is fine);
2. the final trained model ready for local inference;
3. a concise README that lets me retrain or fine-tune the network with fresh images.
If you already have experience with logo detection, OCR on curved surfaces, or colour-based sorting, let me know—the project will move much faster with those skills in place.
During training the system must learn to:
• recognise the disc itself in any orientation, even when it fills only part of the frame;
• focus on brand logos first, using them as the primary classification key;
• capture colour and pattern data so the output can include a short textual description of what it sees;
• flag the general condition (new, lightly used, scratched, etc.) based on surface marks;
• parse the stamped text to return the disc name whenever it is legible.
A working solution should come with:
1. the annotated training/validation pipeline (Python preferred; TensorFlow or PyTorch is fine);
2. the final trained model ready for local inference;
3. a concise README that lets me retrain or fine-tune the network with fresh images.
If you already have experience with logo detection, OCR on curved surfaces, or colour-based sorting, let me know—the project will move much faster with those skills in place.