You can deploy this model and start earning money today!

nutchayaporn/
wifi-csi-har-lstm

Name: wifi-csi-har-lstm
Availability: InStock

Views

No views yet

WiFi CSI HAR LSTM

Ready-to-use PyTorch LSTM model for WiFi CSI Human Activity Recognition.

Files

text

wifi-csi-har-lstm/
|-- inference.py
|-- labels.json
|-- model_config.json
|-- requirements.txt
|-- simple_lstm_best.pth
`-- upload_to_hf.py

Model Contract

Model file: simple_lstm_best.pth
Architecture: 2-layer LSTM + hidden2label classifier
Input tensor shape: (batch, 1024, 468)
Single-sample input shape also accepted: (1024, 468)
Output classes:
- standing
- walking
- get_down
- sitting
- get_up
- lying
- no_person

Backend Usage

Install dependencies:

pip install -r requirements.txt

Use the cached predictor in backend code so the model is loaded only once:

python

from inference import get_predictor

predictor = get_predictor()

def classify_csi(csi_array):
    # csi_array can be a Python list, NumPy array, or torch.Tensor.
    # Accepted shapes: (1024, 468) or (batch, 1024, 468).
    return predictor.predict(csi_array)

Example response:

json

{
  "predictions": [
    {
      "class_index": 0,
      "label": "standing",
      "confidence": 0.98,
      "probabilities": {
        "standing": 0.98,
        "walking": 0.01,
        "get_down": 0.0,
        "sitting": 0.0,
        "get_up": 0.0,
        "lying": 0.0,
        "no_person": 0.01
      }
    }
  ]
}

FastAPI Example

python

from fastapi import FastAPI
from pydantic import BaseModel

from inference import get_predictor

app = FastAPI()
predictor = get_predictor()


class PredictRequest(BaseModel):
    csi: list


@app.post("/predict")
def predict(request: PredictRequest):
    return predictor.predict(request.csi)

CLI Test

Run with a zero-filled sample to verify the model can load:

python inference.py

Run with JSON input:

python inference.py --input_json sample.json

nutchayaporn/wifi-csi-har-lstm