16337
942244bd88
- Engine build scripts (FP16/INT8) - Benchmark validation scripts - Result parsing and analysis tools - COCO dataset configuration
95 lines
2.5 KiB
Batchfile
95 lines
2.5 KiB
Batchfile
@echo off
|
|
chcp 65001 >nul
|
|
echo ============================================
|
|
echo YOLO INT8 Quantization Complete Workflow
|
|
echo ============================================
|
|
echo.
|
|
|
|
REM 激活虚拟环境
|
|
echo [1/4] Activating yolo virtual environment...
|
|
call conda activate yolo
|
|
echo.
|
|
|
|
REM 步骤1: 检查并导出ONNX模型
|
|
echo [2/4] Checking ONNX model...
|
|
if not exist yolo11n.onnx (
|
|
echo ONNX model not found. Exporting from yolo11n.pt...
|
|
if not exist yolo11n.pt (
|
|
echo ERROR: yolo11n.pt not found!
|
|
echo Please place yolo11n.pt in the current directory.
|
|
pause
|
|
exit /b 1
|
|
)
|
|
python -c "from ultralytics import YOLO; model = YOLO('yolo11n.pt'); model.export(format='onnx', dynamic=True, simplify=True, opset=12)"
|
|
if not exist yolo11n.onnx (
|
|
echo ERROR: ONNX export failed!
|
|
pause
|
|
exit /b 1
|
|
)
|
|
echo ONNX model exported successfully.
|
|
) else (
|
|
echo ONNX model already exists: yolo11n.onnx
|
|
)
|
|
echo.
|
|
|
|
REM 步骤2: 生成校准缓存
|
|
echo [3/4] Generating calibration cache...
|
|
if not exist yolo11n_int8.cache (
|
|
echo Running calibration_gen.py...
|
|
python calibration_gen.py
|
|
if not exist yolo11n_int8.cache (
|
|
echo ERROR: Calibration cache generation failed!
|
|
pause
|
|
exit /b 1
|
|
)
|
|
echo Calibration cache generated: yolo11n_int8.cache
|
|
) else (
|
|
echo Calibration cache already exists: yolo11n_int8.cache
|
|
)
|
|
echo.
|
|
|
|
REM 步骤3: 构建TensorRT Engine
|
|
echo [4/4] Building TensorRT Engine...
|
|
echo.
|
|
echo Running trtexec with INT8 calibration...
|
|
echo.
|
|
|
|
trtexec ^
|
|
--onnx=yolo11n.onnx ^
|
|
--saveEngine=yolo11n_int8_b1_8.engine ^
|
|
--explicitBatch ^
|
|
--int8 ^
|
|
--fp16 ^
|
|
--workspace=4096 ^
|
|
--builderOptimizationLevel=5 ^
|
|
--profilingVerbosity=detailed ^
|
|
--calib=yolo11n_int8.cache ^
|
|
--minShapes=input:1x3x640x640 ^
|
|
--optShapes=input:4x3x640x640 ^
|
|
--maxShapes=input:8x3x640x640 ^
|
|
--useCudaGraph ^
|
|
--useSpinWait ^
|
|
--noTF32
|
|
|
|
if %errorlevel% equ 0 (
|
|
echo.
|
|
echo ============================================
|
|
echo TensorRT Engine built successfully!
|
|
echo Output: yolo11n_int8_b1_8.engine
|
|
echo ============================================
|
|
echo.
|
|
echo Next steps:
|
|
echo 1. Run: python quantize_yolo.py
|
|
echo to validate models and calculate mAP drop
|
|
echo.
|
|
) else (
|
|
echo.
|
|
echo ============================================
|
|
echo Engine build failed!
|
|
============================================
|
|
pause
|
|
exit /b 1
|
|
)
|
|
|
|
pause
|