heal
horizon
File size: 2,634 Bytes
e3d5217
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
license: other
tags:
  - heal
  - horizon
---

# EfficientNet-B0

EfficientNet-B0 consists of MBConv (mobile inverted bottleneck) convolution blocks, scaled jointly in depth/width/resolution via compound scaling; this config uses ReLU activation and disables SE Block (better suited for BPU quantization deployment).

---

## Deployment Metrics

### Model Parameters

| Model | Model Input | Backbone | Neck | Model Output |
|---|---|---|---|---|
| EfficientNet-B0 | `1x3x224x224` | EfficientNet-B0 | — | Classification logits `(B,1000)` |

### Accuracy Metrics

| March | Metric | float | calibration | qat | hbm |
| --- | --- | --- | --- | --- | --- |
| J6M | Accuracy | 0.7491 | 0.7433 | — | 0.7436 |
|  | TopKAccuracy(5) | — | — | — | — |

> Results are based on `march = March.NASH_M` (J6M) configuration; this task has no QAT stage (qat column is `—`).
>
> HEAL version: heal 0.0.2 / hbdk4-compiler 4.11.11 / horizon_plugin_pytorch 3.3.10.

### Performance Metrics

> **Performance measurement**: FPS is measured with single-core eight-thread; Latency is measured with single-core single-thread; Memory is peak DDR usage.

| March | latency (ms) | fps | Memory Usage |
|---|---|---|---|
| J6M | 0.40 | 4938.45 | 8.90 |
| J6P | 0.35 | 10480.93 | 9.00 |
| J6B | - | - | - |

J6B performance is not available for this model.

---

## Model Overview

### Core Design

EfficientNet-B0 consists of MBConv (mobile inverted bottleneck) convolution blocks, scaled jointly in depth/width/resolution via compound scaling; this config uses ReLU activation and disables SE Block (better suited for BPU quantization deployment).

- **Task type**: Image classification (Image Classification).
- **backbone**: EfficientNet-B0 (`efficientnet`, `model_type="b0"`, `activation="relu"`, `use_se_block=False`, `num_classes=1000`), composed of MBConv blocks scaled via compound scaling in depth/width/resolution; this config uses ReLU activation and disables SE Block (better suited for BPU quantization deployment).
- **neck**: — (EfficientNet-B0 has built-in fully-connected classification head; no separate neck).
- **Classification head**: EfficientNet-B0 built-in fully-connected classification head, directly outputs 1000-class logits.
- **Loss**: `CEWithLabelSmooth` (cross-entropy with label smoothing).
- **Model input**: Single RGB image, resolution `224 × 224` (`1x3x224x224`).
- **Model output**: 1000-class prediction logits; argmax gives predicted class.

### Official Repo and Paper

Official repo: https://github.com/tensorflow/tpu/tree/master/models/official/efficientnet
Paper: https://arxiv.org/abs/1905.11946