PRECOG-SENSE: Camera-Only Radar Health Prediction

Author: Nikhil Upadhyay | MSc Business Analytics | Dublin Business School Project: PRECOG-AV

Overview

SENSE predicts radar sensor availability from ViT-B/16 camera features alone. It is Module 1 of the PRECOG danger anticipation system for autonomous vehicles.

Results

Metric	Value
Test AUC	1.0000
Test Accuracy	99.71%
Val AUC	0.9649
Parameters	541,569

Trained on 276,445 clips across 20 countries. Tested on Greece and Bulgaria — countries never seen during training. Zero geographic leakage confirmed.

Usage

import torch
import torch.nn as nn
from huggingface_hub import hf_hub_download

class SENSEModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.net = nn.Sequential(
            nn.Linear(768, 512), nn.GELU(), nn.Dropout(0.3),
            nn.Linear(512, 256), nn.GELU(), nn.Dropout(0.2),
            nn.Linear(256, 64),  nn.GELU(), nn.Linear(64, 1))
    def forward(self, x):
        return torch.sigmoid(self.net(x)).squeeze(-1)

path = hf_hub_download("Trazemag/PRECOG-SENSE", "sense_best.pt")
model = SENSEModel()
model.load_state_dict(torch.load(path, map_location="cpu"))
model.eval()
# Input:  ViT-B/16 feature vector (768-dim)
# Output: scalar — near 1.0 = radar present, near 0.0 = radar absent

Citation

@misc{upadhyay2026precog,
  title  = {PRECOG: Proactive Risk and Environmental Cognition for Autonomous Vehicles},
  author = {Upadhyay, Nikhil},
  year   = {2026},
  url    = {https://github.com/TrazeMaG/PRECOG-AV}
}