Skip to content

TinyMyo: Update hand gesture classification experiment details #8

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
MatteoFasulo wants to merge 2 commits into
base: main
Choose a base branch
from
Open

TinyMyo: Update hand gesture classification experiment details #8

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
9d3334c
Update hand gesture classification experiment details
MatteoFasulo Feb 27, 2026
28a362e
Clarify dataset and performance metrics in TinyMyo.md
MatteoFasulo Feb 27, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
24 changes: 12 additions & 12 deletions docs/model/TinyMyo.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -113,9 +113,9 @@ TinyMyo supports three major categories:

Evaluated on:

* **Ninapro DB5** (52 classes, 10 subjects)
* **EPN-612** (5 classes, 612 subjects)
* **UCI EMG** (6 classes, 36 subjects)
* **Ninapro DB5** (52 classes, 10 subjects, 200 Hz)
* **EPN-612** (5 classes, 612 subjects, 200 Hz)
* **UCI EMG** (6 classes, 36 subjects, 200 Hz)
* **Generic Neuromotor Interface** (Meta wristband; 9 gestures)
* Repository: [MatteoFasulo/generic-neuromotor-interface](https://github.com/MatteoFasulo/generic-neuromotor-interface)

Expand All @@ -126,8 +126,8 @@ Evaluated on:
* EMG filtering: **20–90 Hz** bandpass + 50 Hz notch
* Windows:

* **200 ms** (best for DB5)
* **1000 ms** (best for EPN & UCI)
* **1 sec** (best for DB5)
* **5 sec** (best for EPN & UCI)
* Per-channel z-scoring
* Linear classification head

Expand All @@ -138,9 +138,9 @@ Evaluated on:

| Dataset | Metric | Result |
| ------------------------ | -------- | ----------------- |
| **Ninapro DB5 (200 ms)** | Accuracy | **89.41 ± 0.16%** |
| **EPN-612 (1000 ms)** | Accuracy | **96.74 ± 0.09%** |
| **UCI EMG (1000 ms)** | Accuracy | **97.56 ± 0.32%** |
| **Ninapro DB5 (1 sec)** | Accuracy | **89.41 ± 0.16%** |
| **EPN-612 (5 sec)** | Accuracy | **96.74 ± 0.09%** |
| **UCI EMG (5 sec)** | Accuracy | **97.56 ± 0.32%** |
| **Neuromotor Interface** | CLER | **0.153 ± 0.006** |

TinyMyo achieves **state-of-the-art** on DB5, EPN-612, and UCI.
Expand All @@ -149,9 +149,9 @@ TinyMyo achieves **state-of-the-art** on DB5, EPN-612, and UCI.

### **4.2 Hand Kinematic Regression**

Dataset: **Ninapro DB8**
Dataset: **Ninapro DB8** (2000 Hz)
Task: Regress **5 joint angles (DoA)**
Preprocessing: z-score only; windows of **200 ms** or **1000 ms**
Preprocessing: z-score only; windows of **100 ms** or **500 ms**

**Regression head (788k params)**

Expand All @@ -162,7 +162,7 @@ Preprocessing: z-score only; windows of **200 ms** or **1000 ms**

**Performance (Fine-tuned)**

* **MAE = 8.77 ± 0.12°** (1000 ms window)
* **MAE = 8.77 ± 0.12°** (500 ms window)

Although previous works achieve lower MAE (≈6.89°), those models are **subject-specific**, whereas TinyMyo trains **one model across all subjects**, a significantly harder problem.

Expand Down Expand Up @@ -219,7 +219,7 @@ Key elements:
* Integer softmax, integer LayerNorm, integer GELU
* Static liveness-based memory arena

**Runtime (NinaPro EPN612 pipeline):**
**Runtime (EPN612 dataset):**

* **0.785 s inference time**
* **44.91 mJ energy**
Expand Down