yolox_backbone is a deep-learning library and is a collection of YOLOX Backbone models.

Last update: Dec 28, 2022

Overview

YOLOX-Backbone

yolox-backbone is a deep-learning library and is a collection of YOLOX backbone models.

Install

pip install yolox-backbone

Load a Pretrained Model

Pretrained models can be loaded using yolox_backbone.create_model.

import yolox_backbone

m = yolox_backbone.create_model('yolox-s', pretrained=True)
m.eval()

List Supported Models

import yolox_backbone
from pprint import pprint

model_names = yolox_backbone.list_models()
pprint(model_names)

>>> ['yolox-s',
 'yolox-m',
 'yolox-l',
 'yolox-x',
 'yolox-nano',
 'yolox-tiny',
 'yolox-darknet53']

Select specific feature levels

There is one creation argument impacting the output features.

out_features selects which FPN features to output

Example

import yolox_backbone
import torch
from pprint import pprint

pprint(yolox_backbone.list_models())

model_names = yolox_backbone.list_models()
for model_name in model_names:
    print("model_name: ", model_name)
    model = yolox_backbone.create_model(model_name=model_name, 
                                        pretrained=True, 
                                        out_features=["P3", "P4", "P5"]
                                        )

    input_tensor = torch.randn((1, 3, 640, 640))
    fpn_output_tensors = model(input_tensor)

    p3 = fpn_output_tensors["P3"]
    p4 = fpn_output_tensors["P4"]
    p5 = fpn_output_tensors["P5"]
    
    print("input_tensor.shape: ", input_tensor.shape)
    print("p3.shape: ", p3.shape)
    print("p4.shape: ", p4.shape)
    print("p5.shape: ", p5.shape)
    print("-" * 50)

Output:

['yolox-s', 'yolox-m', 'yolox-l', 'yolox-x', 'yolox-nano', 'yolox-tiny', 'yolox-darknet53']
model_name:  yolox-s
input_tensor.shape:  torch.Size([1, 3, 640, 640])
p3.shape:  torch.Size([1, 128, 80, 80])
p4.shape:  torch.Size([1, 256, 40, 40])
p5.shape:  torch.Size([1, 512, 20, 20])
--------------------------------------------------
model_name:  yolox-m
input_tensor.shape:  torch.Size([1, 3, 640, 640])
p3.shape:  torch.Size([1, 192, 80, 80])
p4.shape:  torch.Size([1, 384, 40, 40])
p5.shape:  torch.Size([1, 768, 20, 20])
--------------------------------------------------
model_name:  yolox-l
input_tensor.shape:  torch.Size([1, 3, 640, 640])
p3.shape:  torch.Size([1, 256, 80, 80])
p4.shape:  torch.Size([1, 512, 40, 40])
p5.shape:  torch.Size([1, 1024, 20, 20])
--------------------------------------------------
model_name:  yolox-x
input_tensor.shape:  torch.Size([1, 3, 640, 640])
p3.shape:  torch.Size([1, 320, 80, 80])
p4.shape:  torch.Size([1, 640, 40, 40])
p5.shape:  torch.Size([1, 1280, 20, 20])
--------------------------------------------------
model_name:  yolox-nano
input_tensor.shape:  torch.Size([1, 3, 640, 640])
p3.shape:  torch.Size([1, 64, 80, 80])
p4.shape:  torch.Size([1, 128, 40, 40])
p5.shape:  torch.Size([1, 256, 20, 20])
--------------------------------------------------
model_name:  yolox-tiny
input_tensor.shape:  torch.Size([1, 3, 640, 640])
p3.shape:  torch.Size([1, 96, 80, 80])
p4.shape:  torch.Size([1, 192, 40, 40])
p5.shape:  torch.Size([1, 384, 20, 20])
--------------------------------------------------
model_name:  yolox-darknet53
input_tensor.shape:  torch.Size([1, 3, 640, 640])
p3.shape:  torch.Size([1, 128, 80, 80])
p4.shape:  torch.Size([1, 256, 40, 40])
p5.shape:  torch.Size([1, 512, 20, 20])
--------------------------------------------------

You might also like...

CSWin Transformer: A General Vision Transformer Backbone with Cross-Shaped

CSWin-Transformer This repo is the official implementation of "CSWin Transformer: A General Vision Transformer Backbone with Cross-Shaped Windows". Th

409 Jan 6, 2023

Inflated i3d network with inception backbone, weights transfered from tensorflow

I3D models transfered from Tensorflow to PyTorch This repo contains several scripts that allow to transfer the weights from the tensorflow implementat

479 Dec 8, 2022

PyTorch Implementation of Backbone of PicoDet

PicoDet-Backbone PyTorch Implementation of Backbone of PicoDet Original Implementation is implemented on PaddlePaddle. Example picodet_l_backbone = ES

7 Jul 12, 2022

Adds timm pretrained backbone to pytorch's FasterRcnn model

timmFasterRcnn model_config.py - it returns the model,feat_sizes,output channel and the feat layer names, which is reqd by the Add_FPN.py file Add_FP

12 Dec 3, 2022

An example to implement a new backbone with OpenMMLab framework.

Backbone example on OpenMMLab framework English | 简体中文 Introduction This is an template repo about how to use OpenMMLab framework to develop a new bac

22 Dec 29, 2022

YOLOX is a high-performance anchor-free YOLO, exceeding yolov3~v5 with ONNX, TensorRT, ncnn, and OpenVINO supported.

Introduction YOLOX is an anchor-free version of YOLO, with a simpler design but better performance! It aims to bridge the gap between research and ind

7.7k Jan 3, 2023

using yolox+deepsort for object-tracker

YOLOX_deepsort_tracker yolox+deepsort实现目标跟踪最新的yolox尝尝鲜~~（yolox正处在频繁更新阶段，因此直接链接yolox仓库作为子模块） Install Clone the repository recursively: git clone --rec

245 Dec 26, 2022

MegEngine implementation of YOLOX

Introduction YOLOX is an anchor-free version of YOLO, with a simpler design but better performance! It aims to bridge the gap between research and ind

77 Nov 22, 2022

YOLOX + ROS(1, 2) object detection package

158 Dec 21, 2022

Comments

Support for different number of input channels

Hi there, thanks for making this lightweight package of YOLOX!

I was wondering if the create_model function could be modified to support more inputs channels? At the moment it is hardcoded to support only 3 channels (RGB), but I have a project that will have 4 (or more) channel inputs (e.g. RGBA). Scanning through the code, I think it would require modifying the line below (and tracing the 3 up a few classes back to create_model)

https://github.com/developer0hye/YOLOX-Backbone/blob/681752763ca7dd4db4ecc18693fec1c6cd2e34b2/yolox_backbone/models/darknet.py#L115

Happy to submit a PR if it helps, I just wanted to check if this repo is meant to follow the original YOLOX implementation, or if there is room to expand some more :grin:

opened by weiji14 8