| --- |
| 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 (MB) | |
| |---|---|---|---| |
| | 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 |
| |
