计划执行与检查节点
executing-plans
按既定实施计划完成工作,保持过程检查和反馈。
int8-benchmarking【效率与协作】Routes INT8 calibration, calibration-image preparation, and benchmark tuning tasks for DeepStream-Yolo models.
独立 GitHub 快照 · 自动分类 · 以原作者文档为准
Use this sub-skill when the task is about calibration-enabled builds, calib.table, INT8 engine generation, or interpreting the benchmark notes and NMS values.
calib.tableINT8_CALIB_IMG_PATHINT8_CALIB_BATCH_SIZEOPENCV=1int8-calib-file and related DeepStream config edits.model-conversion.deployment.multi-gie.references/workflows.md for the calibration flow.references/benchmark-notes.md for the benchmark table and NMS defaults.scripts/make-calibration-list.sh to build calibration.txt from a directory of images.scripts/build-nvdsinfer-plugin.sh --output-dir ./deepstream-yolo-runtime to rebuild the library with OPENCV=1 when calibration support is required.deployment only after the calibration build path is ready.references/troubleshooting.md if the calibration build or runtime fails.OPENCV=1.calib.table and uses the INT8 engine path.libopencv-dev missingOPENCV=1 build path not enabledINT8_CALIB_IMG_PATH points to a stale listcalib.table is not writtenscripts/make-calibration-list.sh — build calibration.txt from a directory of images.scripts/build-nvdsinfer-plugin.sh — build the custom library in a fresh runtime tree with OPENCV=1 when needed.references/workflows.md — calibration and run steps.references/benchmark-notes.md — benchmark and NMS guidance.references/troubleshooting.md — calibration-specific failure modes.