Projects

Machine learning and computer vision projects.

interactive_demo

demo_similarity_title

demo_similarity_sub

Query: tabby_cat.jpg → Nearest: 'a photo of a cat' (cos=0.97) · persian.jpg (cos=0.95) · siamese.jpg (cos=0.93)
BENCHMARK LAB · TENSORRT FP16 / CUDA

Precision vs. Inference Speed

Compare production runtime configurations on a 1080p multi-object detection & tracking stream (NVIDIA T4 / A10 GPU).

p99 Frame Latency11.4 ms
Sustained Throughput87.7 FPS
Device Memory Footprint2.1 GB · -0.1% mAP@50
Layer-fused TensorRT FP16 engine with dynamic batching.