aicenter

A python based Soft IOC Server for Sample Alignment using a YOLO model.

Installation

python -m venv my-venv
source my-venv/bin/activate
pip install ai-center

To enabled downloading of the latest model file from HuggingFace, install with the [model] extra as follows:

pip install ai-center[model]

Running

To run the EPICS IOC directly, use the following command:

app.epics --device "AIC001" --model "/path/to/model.pt" --video "video_uri"

Command-line Arguments

• --device: The EPICS root name of the device (e.g., AIC001 or AIC002). (Required)
• --model: Path to the YOLO model file. If omitted, the latest model will be downloaded from HuggingFace. The model is about 50 MB in size.
• --video: The URI for the video stream. (e.g., redis://hostname/0030180F06E5:JPG) (Required)
• --confidence: Object detection confidence threshold. (Optional, default: 0.1)
• -v: Enable verbose logging. (Optional)

Currently, only file://, redis:// and http[s]:// video schemes are supported

EPICS Process Variables

• AIC001:enable: Set to 1/0 to Enable/Disable Inference
• AIC001:loop:score: Score of best loop
• AIC001:loop:box: Bounding Box of best loop
• AIC001:loop:valid: 1 if best loop is still valid
• AIC001:loop:id: id of best loop, not used at the moment
• AIC001:crystal:score: Score of best crystal
• AIC001:crystal:box: Bounding Box of best crystal
• AIC001:crystal:valid: 1 if best crystal is still valid
• AIC001:crystal:id: id of best crystal, not used at the moment
• AIC001:pin:score : Score of best pin
• AIC001:pin:box: Bounding Box of best pin
• AIC001:pin:valid: 1 if best pin is still valid
• AIC001:pin:id : ID of best pin, not used at the moment
• AIC001:extra:box: Bounding Box array of extra Crystals (x1, y1, x2, y2) in sequence, length is 4 x number of extra crystals
• AIC001:extra:score: Array of scores of extra crystals
• AIC001:extra:valid: Array of validity of extra crystals, 1 if valid 0 if invalid
• AIC001:extra:id: Corresponding IDs, of extra crystals (not used at the moment)

If you installed the [view] extra, you can also run the stand-alone inference viewer which displays the inference in a window as follows:

app.view --model "/path/to/model.pt" --video "video_uri"

Command-line Arguments

• --model: Path to the YOLO model file. If omitted, the latest model will be downloaded from HuggingFace. The model is about 50 MB in size.
• --video: The URI for the video stream. (e.g., redis://hostname/0030180F06E5:JPG) (Required)
• --confidence: Object detection confidence threshold. (Optional, default: 0.1)
• -v: Enable verbose logging. (Optional)