edgefirst-decoder

High-performance ML model output decoding for object detection and segmentation.

This crate provides efficient post-processing for YOLO and ModelPack model outputs, supporting both floating-point and quantized inference results.

Supported Models

Features

• Quantized decoding - Direct int8/uint8 processing without dequantization overhead
• Configurable NMS - Class-agnostic or class-aware non-maximum suppression
• Batch processing - Efficient handling of batched model outputs
• Builder pattern - Flexible configuration with sensible defaults

Quick Start

use edgefirst_decoder::{DecoderBuilder, DetectBox, Segmentation};

// Build decoder from model config
let decoder = DecoderBuilder::new()
    .with_score_threshold(0.25)
    .with_iou_threshold(0.7)
    .with_config_json_str(model_config_json)
    .build()?;

// Decode quantized model output
let mut detections: Vec<DetectBox> = Vec::with_capacity(100);
let mut masks: Vec<Segmentation> = Vec::with_capacity(100);

decoder.decode_quantized(
    &[model_output.view().into()],
    &mut detections,
    &mut masks,
)?;

// Process results
for det in &detections {
    println!("Class {} at [{:.1}, {:.1}, {:.1}, {:.1}] score={:.2}",
        det.label, det.bbox.xmin, det.bbox.ymin, det.bbox.xmax, det.bbox.ymax, det.score);
}

Low-Level API

For known model types, use the direct decoding functions:

use edgefirst_decoder::yolo::decode_yolo_det;
use edgefirst_decoder::Quantization;

let mut detections = Vec::with_capacity(100);
decode_yolo_det(
    (output_array.view(), Quantization::new(0.012345, 26)),
    0.25,  // score threshold
    0.7,   // IOU threshold
    Some(edgefirst_decoder::configs::Nms::ClassAgnostic),
    &mut detections,
);

Configuration

Decoders can be configured via JSON/YAML matching the model's output specification:

{
  "decoder": "ultralytics",
  "shape": [1, 84, 8400],
  "quantization": [0.012345, 26],
  "normalized": true
}

NMS Modes

• ClassAgnostic - Suppress overlapping boxes regardless of class (default)
• ClassAware - Only suppress boxes with the same class label
• None - Bypass NMS (for models with built-in NMS)

End-to-End Models (YOLO26)

YOLO26 models embed NMS directly in the model architecture (one-to-one matching heads), eliminating the need for external NMS post-processing.

Configure via the decoder_version field in the model config:

{
  "decoder": "ultralytics",
  "decoder_version": "yolo26",
  "shape": [1, 300, 6],
  "quantization": [0.012345, 26],
  "normalized": true
}

When decoder_version is "yolo26", the decoder:

• Bypasses NMS entirely (the nms config field is ignored)
• Expects post-NMS output format: [batch, N, 6+] where columns are [x1, y1, x2, y2, conf, class, ...]
• Supports both detection-only and detection+segmentation variants

For non-end-to-end YOLO26 exports (end2end=false), use decoder_version: "yolov8" with explicit NMS configuration.

Non-End-to-End Mode

Models exported with end2end=false require external NMS, configurable via the nms field:

{
  "decoder": "ultralytics",
  "decoder_version": "yolov8",
  "nms": "class_agnostic",
  "shape": [1, 84, 8400]
}

Proto Mask API

For segmentation models, the decoder provides two APIs for accessing mask prototype data:

• decode_quantized_proto() — returns raw quantized proto data and mask coefficients without materializing pixel masks
• decode_float_proto() — returns float proto data and mask coefficients

These are preferred when passing mask data to GPU rendering pipelines (e.g., ImageProcessor::draw_masks_proto()), as they avoid the CPU cost of materializing full-resolution masks.

// GPU rendering path: decode proto data, pass to GL for fused rendering
let (detections, proto_data) = decoder.decode_quantized_proto(
    &[model_output.view().into()],
)?;

// Pass proto_data directly to GPU for fused mask overlay
processor.draw_masks_proto(&mut frame, &detections, &proto_data)?;

Model Type Variants

The decoder automatically selects the appropriate model type based on the config:

License

Licensed under the Apache License, Version 2.0. See LICENSE for details.