# Gumpest/YOLOv5-Multibackbone-Compression

YOLOv5 Series Multi-backbone(TPH-YOLOv5, Ghostnet, ShuffleNetv2, Mobilenetv3Small, EfficientNetLite, PP-LCNet, SwinTransformer YOLO), Module(CBAM, DCN), Pruning (EagleEye, Network Slimming), Quantization (MQBench) and Deployment (TensorRT, ncnn) Compression Tool Box.

Repository: https://github.com/Gumpest/YOLOv5-Multibackbone-Compression
Canonical: https://ross.abutalabs.com/products/yolov5-multibackbone-compression
Language: Python
License Family: other
Topics: tph-yolov5, ghostnet-yolov5, shufflenetv2-yolov5, mobilenetv3small-yolov5, efficientnetlite-yolov5, eagleeye, mqbench, pplcnet, swin-transformer, network-slimming, cbam, ncnn, tensorrt
Last push: 2022-04-29T19:35:20+00:00

## Health v2 (maintenance only)
Score: 32/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 1772, "days_push": 1587, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1020, forks 197 (observed 2026-08-28T04:03:15.539553+00:00)

## What it is
A YOLOv5-based toolbox for swapping in lightweight or high-accuracy backbones (TPH-YOLOv5, GhostNet, ShuffleNetV2, MobileNetV3-Small, EfficientNetLite, PP-LCNet, Swin Transformer) and applying compression techniques like pruning (EagleEye, Network Slimming) and quantization (MQBench), with deployment support for TensorRT and ncnn. It is primarily demonstrated on the VisDrone drone detection dataset.

## Use cases
- compress a YOLOv5 model for edge deployment
- replace YOLOv5 backbone with a lightweight network like GhostNet or MobileNetV3
- prune a YOLOv5 detection model with EagleEye or network slimming
- quantize YOLOv5 for TensorRT or ncnn inference
- train TPH-YOLOv5 on drone imagery like VisDrone
- build a Swin Transformer YOLOv5 variant for higher accuracy

## When to choose
- you need to shrink YOLOv5 models for mobile or embedded inference
- you want to benchmark multiple lightweight backbones within the YOLOv5 framework
- you are working on small-object detection such as drone-view datasets
- you want a single repo covering pruning, quantization, and deployment for YOLOv5

## When to avoid
- you need a maintained project with active releases and a clear license
- you use detection frameworks other than YOLOv5
- you need production-grade support or commercial licensing
- you want a plug-and-play tool rather than research code requiring manual setup

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, deep-learning, image-processing, computer-vision, benchmarking
- domain: computer-vision, deep-learning, machine-learning, image-processing
- platform: python
- tags: yolov5, object-detection, model-compression, pruning, quantization, tensorrt, ncnn, lightweight-backbones, tph-yolov5, visdrone, linux, gpu

## Member repositories
- Gumpest/YOLOv5-Multibackbone-Compression (main) score 32

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:15.539553+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T07:08:53.587696+00:00, confidence not recorded.
  - readme: https://github.com/Gumpest/YOLOv5-Multibackbone-Compression (fetched 2026-08-28T04:03:15.539553+00:00, sha fa45aa84e3b9)
- Data as of 2026-08-30T08:39:29.467469+00:00.