Point Cloud Object Detection Workflow creation using Pytorch
Budget: $8 – $15 USD
I am starting a series of point-cloud projects and need a senior machine-learning computer vision professional—someone who can prove more than seven years of hands-on work with computer-vision models on Freelancer.
I review only proven working histories, not beautiful bid messages.
Please bid only if your profile already demonstrates substantial point-cloud or 3D deep-learning work.
Your first task is a focused kickoff: build and document a workflow that reliably counts cylinders inside sample point-cloud data.
Here is what I expect:
• Use either Open3D or Torch Points3D (whichever you know best) to set up the end-to-end pipeline—data loading, preprocessing, model selection, training, and inference.
• Deliver the code in a clear, reproducible notebook; Jupyter, Google Colab, or Kaggle are all fine as long as it runs without extra setup.
• The model’s primary metric is an accurate cylinder count. I will provide annotated sample clouds for testing; your output must match the ground-truth totals.
• Comment the critical steps so I can later fine-tune the approach for more complex objects.
A concise, working solution here will lead directly to larger, long-term assignments.
I review only proven working histories, not beautiful bid messages.
Please bid only if your profile already demonstrates substantial point-cloud or 3D deep-learning work.
Your first task is a focused kickoff: build and document a workflow that reliably counts cylinders inside sample point-cloud data.
Here is what I expect:
• Use either Open3D or Torch Points3D (whichever you know best) to set up the end-to-end pipeline—data loading, preprocessing, model selection, training, and inference.
• Deliver the code in a clear, reproducible notebook; Jupyter, Google Colab, or Kaggle are all fine as long as it runs without extra setup.
• The model’s primary metric is an accurate cylinder count. I will provide annotated sample clouds for testing; your output must match the ground-truth totals.
• Comment the critical steps so I can later fine-tune the approach for more complex objects.
A concise, working solution here will lead directly to larger, long-term assignments.