davidr99/locateanything-delivery-detector
LocateAnything delivery-vehicle detector
Drag-and-drop demo for the clean, one-epoch R4 LocateAnything-3B delivery detector. R4 was trained directly from NVIDIA's base model using original-aspect images rather than continuing from the earlier adapters. The app detects Amazon, UPS, USPS-Truck, Other-Vehicles, and FedEx.
The model emits native LocateAnything bounding boxes as normalized integer coordinates in [0, 1000]. The UI also reports normalized [0, 1] and original-image pixel coordinates. Inference uses deterministic decoding so repeated tests on the same image and class prompt are comparable.
Space configuration
- Hardware: ZeroGPU (
zero-a10g) - Model: `davidr99/locateanything-3b-delivery-detector-r4-clean-original-aspect`
- Variable:
MODEL_REPO_ID=davidr99/locateanything-3b-delivery-detector-r4-clean-original-aspect - Optional secret:
HF_TOKENis only needed whenMODEL_REPO_IDis private
Input images are capped at 640 pixels on the longest side for model inference, matching fine-tuning. Boxes are rendered on the original-resolution image.
License
This is a derivative application for NVIDIA LocateAnything-3B and remains restricted to non-commercial research/evaluation under NVIDIA's model license. Keep all required NVIDIA attribution and license notices with the trained model.