Object Detection
Keras
LiteRT
person-detection
search-and-rescue
drone
aerial
tinyml
int8
esp32
centernet
Instructions to use inviolus/VoxHOD with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Keras
How to use inviolus/VoxHOD with Keras:
# !pip install -U keras tensorflow huggingface_hub # Keras needs TensorFlow installed to read "hf://" paths, so the tensorflow backend is selected here; # "jax" and "torch" also work for computation once TensorFlow is installed. import os os.environ["KERAS_BACKEND"] = "tensorflow" import keras model = keras.saving.load_model("hf://inviolus/VoxHOD") - Notebooks
- Google Colab
- Kaggle
Point to the non-commercial VoxHOD-NC branch
Browse files
README.md
CHANGED
|
@@ -14,18 +14,16 @@ tags:
|
|
| 14 |
- centernet
|
| 15 |
---
|
| 16 |
|
| 17 |
-
# VoxHOD: tiny
|
| 18 |
|
| 19 |
A CenterNet-style, anchor-free person detector for downward-looking drone
|
| 20 |
cameras: **136,547 parameters, 197 KB full-int8 TFLite**, or **37,363 parameters,
|
| 21 |
75 KB** for the nano variant, built for ESP32-S3-class microcontrollers. It marks where people are in a 160×160 RGB
|
| 22 |
-
tile so a drone can flag places worth a closer look
|
| 23 |
-
has been trained and measured for so far.
|
| 24 |
|
| 25 |
-
> **Türkçe özet:**
|
| 26 |
-
> çalışacak kadar küçük (136 bin parametre, int8 dosyası 197 KB)
|
| 27 |
-
>
|
| 28 |
-
> ölçüldü. HERIDAL test setinde int8 hâli, eşik 0,2'de insanların
|
| 29 |
> %83,9'unu buluyor. 75 KB'lık nano sürümü %81,2'sini buluyor. Eğitimde
|
| 30 |
> kullanılmayan bir WiSARD klibinde bu oranlar %61 ve %62'ye düşüyor. CC BY 4.0.
|
| 31 |
|
|
@@ -109,6 +107,8 @@ people), never trained on. Treat it as a first signal.
|
|
| 109 |
|
| 110 |
## Data and license
|
| 111 |
|
|
|
|
|
|
|
| 112 |
Trained on [HERIDAL](https://zenodo.org/records/5662351) (CC BY 3.0):
|
| 113 |
D. Božić-Štulić, Ž. Marušić, S. Gotovac, "Deep Learning Approach in Aerial
|
| 114 |
Imagery for Supporting Land Search and Rescue Missions", IJCV 2019.
|
|
@@ -121,7 +121,5 @@ Weights: **CC BY 4.0**. Training code: MIT.
|
|
| 121 |
|
| 122 |
## Intended use
|
| 123 |
|
| 124 |
-
|
| 125 |
-
|
| 126 |
-
lost or injured. Other settings are unmeasured. It is not intended for
|
| 127 |
-
targeting or harming people.
|
|
|
|
| 14 |
- centernet
|
| 15 |
---
|
| 16 |
|
| 17 |
+
# VoxHOD: tiny person detector for drone search and rescue
|
| 18 |
|
| 19 |
A CenterNet-style, anchor-free person detector for downward-looking drone
|
| 20 |
cameras: **136,547 parameters, 197 KB full-int8 TFLite**, or **37,363 parameters,
|
| 21 |
75 KB** for the nano variant, built for ESP32-S3-class microcontrollers. It marks where people are in a 160×160 RGB
|
| 22 |
+
tile so a drone can flag places worth a closer look for a rescue team.
|
|
|
|
| 23 |
|
| 24 |
+
> **Türkçe özet:** Arama kurtarma dronları için, mikrodenetleyicide
|
| 25 |
+
> çalışacak kadar küçük (136 bin parametre, int8 dosyası 197 KB) bir insan
|
| 26 |
+
> tespit modeli. HERIDAL test setinde int8 hâli, eşik 0,2'de insanların
|
|
|
|
| 27 |
> %83,9'unu buluyor. 75 KB'lık nano sürümü %81,2'sini buluyor. Eğitimde
|
| 28 |
> kullanılmayan bir WiSARD klibinde bu oranlar %61 ve %62'ye düşüyor. CC BY 4.0.
|
| 29 |
|
|
|
|
| 107 |
|
| 108 |
## Data and license
|
| 109 |
|
| 110 |
+
**Not trained on the Ferhat Tatık footage.** That footage is licensed to this project for non-commercial use only, so models trained on it are published separately as [VoxHOD-NC](https://huggingface.co/inviolus/VoxHOD-NC) (CC BY-NC 4.0). VoxHOD stays CC BY 4.0.
|
| 111 |
+
|
| 112 |
Trained on [HERIDAL](https://zenodo.org/records/5662351) (CC BY 3.0):
|
| 113 |
D. Božić-Štulić, Ž. Marušić, S. Gotovac, "Deep Learning Approach in Aerial
|
| 114 |
Imagery for Supporting Land Search and Rescue Missions", IJCV 2019.
|
|
|
|
| 121 |
|
| 122 |
## Intended use
|
| 123 |
|
| 124 |
+
Search and rescue: helping find people who are lost or injured. It is not
|
| 125 |
+
intended for targeting or harming people.
|
|
|
|
|
|