Downloads · 30 days
0
PCB-Prune-YOLO-TensorRT-FP16
thangkt/PCB-Prune-YOLO-TensorRT-FP16
PCB-Prune-YOLO-TensorRT-FP16 is a object detection model from thangkt. Use it when you need objects located in an image. It is set up for tensorrt. The card lists the license as agpl-3.0.
Static batch-1 TensorRT FP16 engines for the baseline and direct DepGraph P10, P20, and P30 checkpoints. Engines were built and benchmarked directly on Tesla T4 with TensorRT 10.16.1.11, CUDA 12.8, PyTorch 2.10.0+cu12…
Access
Public
Updated Aug 2, 2026
Repo size
29.3 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.engine29.3 MB · 100%
At a glance
- Task
- Object Detection
- Library
- tensorrt
- License
- agpl-3.0
- Access
- Public
- Created
- Aug 2, 2026
- Updated
- Aug 2, 2026
- SHA
- 97721f54
Datasets
Model card
From the Hugging Face model README
PCB-Prune-YOLO TensorRT FP16
Static batch-1 TensorRT FP16 engines for the baseline and direct DepGraph P10,
P20, and P30 checkpoints. Engines were built and benchmarked directly on Tesla
T4 with TensorRT 10.16.1.11, CUDA 12.8, PyTorch 2.10.0+cu128, and Ultralytics
8.4.115. Input is [1,3,640,640]; raw output is [1,10,8400]; NMS is external.
| Model | Validation mAP50-95 | Mean latency | FPS | Engine size |
|---|---|---|---|---|
| Baseline | 0.78716 | 1.837 ms | 544.28 | 7.477 MiB |
| P10 direct | 0.77842 | 2.023 ms | 494.34 | 7.627 MiB |
| P20 direct | 0.76931 | 1.933 ms | 517.45 | 7.378 MiB |
| P30 direct | 0.75610 | 1.754 ms | 569.97 | 5.482 MiB |
Latency is pure synchronized engine forward after 50 warm-ups across 200 measurements and excludes preprocessing/NMS. TensorRT engines are not portable across arbitrary TensorRT/CUDA/GPU combinations; rebuild them on the deployment target when the stack differs. Full provenance and reports are included.
- Task
- Object Detection
- Library
- tensorrt
- License
- agpl-3.0
- Created
- Aug 2, 2026
- Updated
- Aug 2, 2026