edgefirst-decoder 0.22.2

ML model output decoding for YOLO and ModelPack object detection and segmentation
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96

// SPDX-FileCopyrightText: Copyright 2026 Au-Zone Technologies
// SPDX-License-Identifier: Apache-2.0

//! Cross-repo contract test for DE-2651: build a [`Decoder`] from the
//! schema-v2 `edgefirst.json` that ModelPack now embeds for a multitask
//! (detection + segmentation) model.
//!
//! ModelPack emits, in order: per-FPN-scale `type: detection` anchor grids,
//! a raw `type: segmentation` per-class map (canonical NHWC `[1, H, W,
//! num_classes]`), and a decoded `type: masks` argmax class-index map
//! (`[1, H, W]`). HAL must classify this as `ModelPackSegDetSplit` — the
//! decoded `masks` head is not consumed by the ModelPack decoder path and must
//! not prevent the build, and the segmentation class count is taken from the
//! NHWC channel axis.

use edgefirst_decoder::{schema::SchemaV2, DecoderBuilder};

/// A 2-scale multitask ModelPack model with 2 annotated classes.
/// Detection: 3 anchors × (5 + 2 classes) = 21 channels per grid cell.
/// Segmentation: 2 classes + background = 3 channels (NHWC).
const MODELPACK_MULTITASK_JSON: &str = r#"
{
  "schema_version": 2,
  "input": { "shape": [1, 3, 480, 640] },
  "dataset": { "classes": ["cup", "saucer"] },
  "outputs": [
    {
      "name": "output_0", "type": "detection", "encoding": "anchor",
      "decoder": "modelpack", "normalized": true,
      "shape": [1, 30, 40, 21],
      "dshape": [{"batch": 1}, {"height": 30}, {"width": 40},
                 {"num_anchors_x_features": 21}],
      "stride": [16, 16],
      "anchors": [[0.05, 0.05], [0.1, 0.1], [0.2, 0.2]],
      "dtype": "float32", "quantization": null
    },
    {
      "name": "output_1", "type": "detection", "encoding": "anchor",
      "decoder": "modelpack", "normalized": true,
      "shape": [1, 15, 20, 21],
      "dshape": [{"batch": 1}, {"height": 15}, {"width": 20},
                 {"num_anchors_x_features": 21}],
      "stride": [32, 32],
      "anchors": [[0.3, 0.3], [0.5, 0.5], [0.8, 0.8]],
      "dtype": "float32", "quantization": null
    },
    {
      "name": "output_seg", "type": "segmentation", "decoder": "modelpack",
      "shape": [1, 480, 640, 3],
      "dshape": [{"batch": 1}, {"height": 480}, {"width": 640},
                 {"num_classes": 3}],
      "dtype": "float32", "quantization": null
    },
    {
      "name": "output_seg_decoded", "type": "masks", "decoder": "modelpack",
      "shape": [1, 480, 640],
      "dshape": [{"batch": 1}, {"height": 480}, {"width": 640}],
      "dtype": "int32", "quantization": null
    }
  ]
}
"#;

#[test]
fn build_decoder_from_modelpack_multitask_schema() {
    let schema = SchemaV2::parse_json(MODELPACK_MULTITASK_JSON)
        .expect("parse ModelPack multitask edgefirst.json");
    // The decoded `masks` head must not prevent the build: HAL ignores it in
    // the ModelPack path and decodes the raw `segmentation` head. Previously
    // an untagged/mis-tagged or undeclared output produced
    // InvalidConfig("Invalid ModelPack model outputs") (DE-2651).
    DecoderBuilder::new()
        .with_schema(schema)
        .build()
        .expect("build ModelPack multitask decoder");
}

#[test]
fn parse_tolerates_unknown_input_dshape_axis_name() {
    // Older ModelPack exports tagged the input dshape channel axis as
    // `channels`, which HAL's DimName enum does not define. Parsing must not
    // reject the whole metadata over an unrecognised axis name — the unknown
    // axis is preserved as `DimName::Unknown` and sorts to the canonical tail
    // (DE-2651).
    let json = MODELPACK_MULTITASK_JSON.replace(
        "\"input\": { \"shape\": [1, 3, 480, 640] }",
        "\"input\": { \"shape\": [1, 480, 640, 3], \"dshape\": [\
         {\"batch\": 1}, {\"height\": 480}, {\"width\": 640}, {\"channels\": 3}] }",
    );
    let schema =
        SchemaV2::parse_json(&json).expect("parse metadata with unknown input dshape axis name");
    DecoderBuilder::new()
        .with_schema(schema)
        .build()
        .expect("build decoder despite unknown input dshape axis name");
}