Apppublic

lakxjass/recursive-sampler-214

sourceHugging Faceapache-2.0updated 3mo agoView on Hugging Face

0likes

App README

EfficientNet-CNN-Classifier

A lightweight image classification runtime optimized for edge deployment and batch inference workloads. Built on ONNX Runtime with a custom CNN backbone trained on a proprietary dataset.

Features

—ONNX Runtime inference — CPU-optimized, no GPU required
—Batch processing — configurable concurrency for high-throughput workloads
—Edge-friendly — minimal memory footprint, runs on 2 vCPU instances
—REST API — simple HTTP interface for integration with downstream pipelines

API

Health Check

GET /health → {"status": "ok", "uptime": 12345}

Batch Inference

POST /api/v1/infer
Content-Type: application/json

{
  "batch_size": 5,
  "workers": 2,
  "params": { "code": "...", "retries": 2 }
}

Model Details

Property	Value
Framework	ONNX Runtime
Architecture	Custom CNN (8 layers)
Input size	160×48 RGB
Output	5-class softmax
Parameters	~2.1M
Quantization	INT8 (planned)

Environment Variables

Variable	Default	Description
`APP_MAX_WORKERS`	`2`	Max concurrent inference workers
`APP_TASK_TIMEOUT`	`180`	Per-task timeout (seconds)
`PORT`	`7860`	Server listen port

License

Apache 2.0