GPU-resident TensorRT pipeline for FoundationStereo. 2.7x faster on Jetson Orin, bit-identical FP16, <0.2% INT8 error.
cuda jetson stereo-matching depth-estimation tensorrt int8-quantization jetson-orin realtime-inference foundationstereo flashstereo
-
Updated
May 26, 2026 - Python