ti-edgeai-generate-config
Build, validate and run edgeai-gst-apps YAML configs (inputs, models, outputs, flows) on TI Edge AI boards with a C7x/MMA accelerator (TDA4VM, AM68A, AM69A, AM67A, AM62A): video file, image sequence, USB webcam, CSI camera, RTSP or test-pattern input; one or several models; display with FPS overlay, saved video/images, network stream or fakesink; mosaic layouts. Use this skill whenever the user wants to "run my model on the TI board", show live results on a display, use a webcam or camera with the TI demo app, record annotated video, stream detections, write an app_config / edgeai-gst-apps yaml, or asks how the TI GStreamer pipeline (tiovxmultiscaler, tiovxdlpreproc, kmssink) is assembled - even if they never say "config".
- License 许可
- MIT
- 最近更新
- 2026-10-02
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- ti-edgeai-generate-config
- license
- MIT
- metadata
- { "version": "0.1.0", "tested_on": "TDA4VM, Edge AI SDK 11.0: image, video, classification, two-model mosaic and USB-webcam-to-file runs. Display output not confirmed" }
- description
- Build, validate and run edgeai-gst-apps YAML configs (inputs, models, outputs, flows) on TI Edge AI boards with a C7x/MMA accelerator (TDA4VM, AM68A, AM69A, AM67A, AM62A): video file, image sequence, USB webcam, CSI camera, RTSP or test-pattern input; one or several models; display with FPS overlay, saved video/images, network stream or fakesink; mosaic layouts. Use this skill whenever the user wants to "run my model on the TI board", show live results on a display, use a webcam or camera with the TI demo app, record annotated video, stream detections, write an app_config / edgeai-gst-apps yaml, or asks how the TI GStreamer pipeline (tiovxmultiscaler, tiovxdlpreproc, kmssink) is assembled - even if they never say "config".
02 资源文件 16
- 参考文档agents/openai.yaml412 B
- 资产文件assets/example_classification.yaml498 B
- 资产文件assets/example_image.yaml524 B
- 资产文件assets/example_mosaic.yaml724 B
- 资产文件assets/example_video.yaml537 B
- 资产文件assets/example_webcam_display.yaml588 B
- 资产文件assets/example_webcam.yaml584 B
- 参考文档evals/evals.json4.8 KB
- 参考文档references/camera-and-display.md1.9 KB
- 参考文档references/config-schema.md2.7 KB
- 参考文档references/pipeline-anatomy.md1.5 KB
- 脚本scripts/discover_cameras.sh1.1 KB
- 脚本scripts/generate_config.py7.2 KB
- 脚本scripts/validate_config.py9.5 KB
- 脚本tests/test_config_tools.py8.7 KB
- 脚本tests/test_discover_cameras.py2.1 KB
03 SKILL.md 正文
Configs and Runs for edgeai-gst-apps
When this skill is active, read references/config-schema.md before writing a config by hand and run scripts/validate_config.py
before touching the board. Prerequisites and shared rules (SoC key, [H, W] order, display ownership): ti-edgeai-dev. The model
folder must already exist on the board (ti-edgeai-import-model).
Principles
- Minimal first. Build the smallest working config (one input, one model, one output) and add mosaics, overlays or extra flows only when asked.
- Finite input for the first run (image sequence or video file, file output): it ends by itself and can be checked.
- Look at the output. "exit 0" is not a result; open a frame and check boxes, aspect and labels.
- Say what ran. Report the input kind, config, log evidence, and what was not confirmed.
Workflow
- Collect requirements (infer from the request, ask only for the unknowns): input kind and its real mode, model folder(s),
output kind, display size, how long to run.
- Camera: discover it, do not assume a node number:
BOARD=root@<board-ip> scripts/discover_cameras.sh(capture nodes, card names, formats and sizes; codec nodes skipped) and choose a mode the camera really offers. Prefer the mode whose aspect matches the model input (16:9 -> 640x360 for a 640x352 model). Codec nodes (hardware decoder/encoder) are not cameras; the discovery script skips them. YUYV-only webcams needformat: auto(notjpeg). - Output size must be <= input size (the app exits otherwise).
- Camera: discover it, do not assume a node number:
- Generate:
python3 scripts/generate_config.py --input usb --source /dev/video<N> --width 640 --height 360 --format auto --model /opt/model_zoo/<model> --viz-threshold 0.3 --output display --perf-overlay graph -o my.yaml(several--modelflags tile them with mosaic rectangles). Edit by hand for anything the generator does not expose;references/config-schema.mdlists every key. - Validate:
python3 scripts/validate_config.py my.yaml --board root@<board-ip>(schema, name references, size/mosaic rules, and a read-only SSH check that each model folder has param.yaml, dataset.yaml and a compiled net.bin). Fix every ERROR. - Install and run, look at the output, report what you saw.
Running
Finite input, saved output (the safe first test; GST_SOC is the SoC key from ti-edgeai-dev/references/platforms.md):
scp my.yaml root@<board-ip>:/opt/edgeai-gst-apps/configs/
ssh root@<board-ip> 'cd /opt/edgeai-gst-apps/apps_python && export SOC=<GST_SOC> TERM=xterm &&
timeout 120 ./app_edgeai.py ../configs/my.yaml > /tmp/run.log 2>&1; echo exit=$?;
grep -aE "VX_ZONE_ERROR\]|rror|Fail" /tmp/run.log | grep -v -e curses -e setupterm'
Success: exit=0, Offloaded Nodes - N, Total Nodes - N, output files present (ls /opt/edgeai-test-data/output). Then scp a frame
back and look at it. Delete test outputs afterwards (small disk).
Camera to saved frames: the same with timeout -s INT 25 (cameras never finish; exit 124 is normal).
Live on a display. This changes the board's system state (the stock GUI is stopped and a transient unit is started), so:
- Get explicit authorization first. Say exactly what you will do and what the user will lose on screen, and wait for a yes.
Do not stop
edgeai-initon inference from an earlier request such as "show it on the display". The user must also be present to watch. - Record the original state and keep it in your report:
systemctl is-active edgeai-init; systemctl is-enabled edgeai-init(and whetheredgeai-demoalready exists). - Run (only after the yes):
ssh root@<board-ip> 'systemctl stop edgeai-init' # the stock GUI owns the screen
ssh root@<board-ip> 'systemd-run --unit=edgeai-demo --setenv=SOC=<GST_SOC> --setenv=TERM=xterm \
--setenv=PYTHONPATH=/usr/lib/python3.12/site-packages --working-directory=/opt/edgeai-gst-apps/apps_python \
script -qfc "./app_edgeai.py ../configs/my_display.yaml" /dev/null'
ssh root@<board-ip> 'systemctl is-active edgeai-demo' # confirm it is alive
- Restore on every exit path (done, failed, or the user asks to stop):
systemctl stop edgeai-demo; systemctl start edgeai-init(only if it was active before), then checksystemctl is-active edgeai-initand report the state you left the board in. Why a unit: the status screen needs a tty (scriptprovides one) andnohup &children of an ssh command are killed when it ends. The same consent rule covers firmware updates, reboots and USB/driver resets (ti-edgeai-dev, critical rule 11).
Display status: not confirmed. With the stock GUI running, a display-sink run showed nothing on screen; after stopping it the
app ran as a unit without errors, but a picture was never confirmed. Report display runs as unconfirmed until the user says they
see video; if the screen stays blank check /sys/class/drm/card*-*/status and the connector: option
(references/camera-and-display.md).
Pipeline sanity
With the right SOC the app builds tiovxdlcolorconvert ! tiovxmultiscaler -> {full-res branch, tiovxdlpreproc branch}. Print it:
grep -a "multifilesrc\|v4l2src\|tiovx" /tmp/run.log | cut -c1-300. videoscale ! videoconvert means the ARM-only path (missing SOC).
Details: references/pipeline-anatomy.md.
Reference Documents
| File | Use when |
|---|---|
| references/config-schema.md | every config key and rule |
| references/pipeline-anatomy.md | elements, tracers, OpTIFlow, C++ app |
| references/camera-and-display.md | device discovery, formats, display ownership, USB failure modes |
Files
scripts/generate_config.py,scripts/validate_config.py: PyYAML-serialized output; validator checks types and ranges (unit-tested).scripts/discover_cameras.sh: read-only V4L2 capture-device discovery (unit-tested with a simulated v4l2-ctl; not run on a board).assets/example_{image,video,classification,mosaic,webcam,webcam_display}.yaml: starting points using TI zoo models (adapt model paths).
04 同仓库更多 Skills
other-name
x
ti-edgeai-train-model
Plan and run training/fine-tuning so a vision model (detector, classifier, segmenter) deploys well on TI Edge AI processors with a C7x/MMA accelerator (TDA4VM, AM68A, AM69A, AM67A, AM62A): pick a TIDL-friendly architecture and input size, choose TI's toolchain (edgeai-modelmaker, edgeai-yolox, edgeai-mmdetection, edgeai-torchvision, plain PyTorch), build datasets and leakage-safe splits, train with quantization-friendly settings, and evaluate with the same preprocessing the board uses before handing the checkpoint to ti-edgeai-import-model. Use this skill whenever the user wants to train, fine-tune, label, split or evaluate a model "for the TI board / edge NPU / C7x", asks which architecture or input size to use, why their detector scores near zero, how to make a model quantization-friendly, or how to prepare labels so the model is learnable - even before any deployment question comes up.
ti-edgeai-profile-pipeline
Measure and improve inference speed on TI Edge AI boards with a C7x/MMA accelerator (TDA4VM, AM68A, AM69A, AM67A, AM62A): model-only latency, per-layer C7x cycle hotspots, per-element GStreamer latency and FPS of the edgeai-gst-apps pipeline, system load, and the levers that trade accuracy for speed (input size, 16-bit layers, frame rate, output size). Use this skill whenever the user asks how fast a model runs on the TI board, wants FPS, latency, "why is it slow", per-layer timings, which stage is the bottleneck, whether the pipeline can keep 30 fps, or how much a precision/size change costs - even if they only say "benchmark it on the TI device".
ti-edgeai-import-model
Get a trained neural network that exports to a static-shape ONNX (a classifier, a segmentation model, or a detector with a TIDL meta-architecture such as YOLOX/SSD/YOLOv5-style) running on a TI Edge AI board's C7x/MMA accelerator (TDA4VM, AM68A, AM69A, AM67A, AM62A): preflight the ONNX, fold input normalisation, compile and int8-calibrate with edgeai-tidl-tools in Docker (tools tag matched to the board SDK), verify offload and accuracy in host emulation, check board-vs-host outputs, package model/ artifacts/ param.yaml dataset.yaml, deploy to /opt/model_zoo and smoke-test with edgeai-gst-apps. Use this skill whenever the user wants to deploy, convert, compile, quantize, port or "put my model on" a TI Jacinto/AM6xA board, mentions TIDL compilation, artifacts, tidl_net.bin, meta_arch_type, a prototxt, calibration images, int8 accuracy loss on the TI device, or asks why a compiled model fails on the board - for any architecture within those contracts, and even if they only say "deploy to the TI device".
ti-edgeai-dev
Texas Instruments Edge AI SDK development for Jacinto 7 / AM6xA processors with a C7x DSP + MMA deep-learning accelerator (TDA4VM, AM68A, AM69A, AM67A, AM62A): TIDL model compilation with edgeai-tidl-tools, the on-board runtime (ONNX Runtime with the TIDL execution provider), edgeai-gst-apps configs and tiovx* GStreamer plugins, model folders and param.yaml, tools-to-SDK version matching, and troubleshooting. Use this skill whenever the user mentions TIDL, C7x, MMA, TDA4VM, J721E, AM68A, AM69A, AM67A, AM62A, edgeai-tidl-tools, edgeai-gst-apps, /opt/model_zoo, tiovxmultiscaler, "TI edge NPU/accelerator", or asks how to run, deploy, debug or speed up a neural network on a TI Jacinto/Sitara board - even if they never say "skill". Start here, then hand off to ti-edgeai-import-model, ti-edgeai-generate-config, ti-edgeai-profile-pipeline or ti-edgeai-train-model.