• 📢 News!
• My column
• Introduction
• All task training results
• Environments
• Download my pretrained models and experiments records
• Prepare datasets
  • CIFAR10
  • CIFAR100
  • ImageNet 1K(ILSVRC2012)
  • ImageNet 21K(Winter 2021 release)
  • ACCV2022
  • VOC2007 and VOC2012
  • COCO2017
  • SAMACOCO
  • Objects365(v2,2020)
  • ADE20K
  • CelebA-HQ
  • FFHQ
• How to train or test a model
• How to use gradio demo
• Reference
• Citation

📢 News!

• 2024/04/15: support segment-anything model training/testing/jupyter example/gradio demo.

My column

https://www.zhihu.com/column/c_1692623656205897728

Introduction

This repository provides simple training and testing examples for following tasks:

task	support dataset	support network
Image classification task	CIFAR100 ImageNet1K(ILSVRC2012) ImageNet21K(Winter 2021 release)	Convformer DarkNet ResNet VAN ViT
Knowledge distillation task	ImageNet1K(ILSVRC2012)	DML loss(ResNet) KD loss(ResNet)
Masked image modeling task	ImageNet1K(ILSVRC2012)	MAE(ViT)
Object detection task	COCO2017 Objects365(v2,2020) VOC2007 and VOC2012	DETR DINO-DETR RetinaNet FCOS
Semantic segmentation task	ADE20K COCO2017	DeepLabv3+
Instance segmentation task	COCO2017	SOLOv2 YOLACT
Salient object detection task	combine dataset	pfan-segmentation
Human matting task	combine dataset	pfan-matting
OCR text detection task	combine dataset	DBNet
OCR text recognition task	combine dataset	CTC Model
Face detection task	combine dataset	RetinaFace
Face parsing task	FaceSynthetics CelebAMask-HQ	pfan-face-parsing
Human parsing task	LIP CIHP	pfan-human-parsing
Interactive segmentation task	combine dataset	SAM(segment-anything)
Diffusion model task	CelebA-HQ CIFAR10 CIFAR100 FFHQ	DDPM DDIM

All task training results

Most experiments were trained on 2-8 RTX4090D GPUs, pytorch2.3, ubuntu22.04.

See all task training results in results.md.

Environments

1、This repository only supports running on ubuntu(verison>=22.04 LTS).

2、This repository only support one node one gpu/one node multi gpus mode with pytorch DDP training.

3、Please make sure your Python environment version>=3.9 and pytorch version>=2.0.

4、If you want to use torch.complie() function,using pytorch2.0/2.2/2.3,don't use pytorch2.1.

Use pip or conda to install those Packages in your Python environment:

torch
torchvision
pillow
numpy
Cython
pycocotools
opencv-python
scipy
einops
scikit-image
pyclipper
shapely
imagesize
nltk
tqdm
yapf
onnx
onnxruntime
onnxsim
thop==0.1.1.post2209072238
gradio==3.50.0
transformers==4.41.2
open-clip-torch==2.24.0

If you want to use xformers,install xformers Packge from offical github repository:

https://github.com/facebookresearch/xformers

If you want to use dino-detr model,install MultiScaleDeformableAttention Packge in your Python environment:

cd to simpleAICV/detection/compile_multiscale_deformable_attention,then run commands:

chmod +x make.sh
./make.sh

Download my pretrained models and experiments records

You can download all my pretrained models and experiments records/checkpoints from huggingface or Baidu-Netdisk.

If you only want to download all my pretrained models(model.state_dict()),you can download pretrained_models folder.

# huggingface
https://huggingface.co/zgcr654321/0.classification_training/tree/main
https://huggingface.co/zgcr654321/1.distillation_training/tree/main
https://huggingface.co/zgcr654321/2.masked_image_modeling_training/tree/main
https://huggingface.co/zgcr654321/3.detection_training/tree/main
https://huggingface.co/zgcr654321/4.semantic_segmentation_training/tree/main
https://huggingface.co/zgcr654321/5.instance_segmentation_training/tree/main
https://huggingface.co/zgcr654321/6.salient_object_detection_training/tree/main
https://huggingface.co/zgcr654321/7.human_matting_training/tree/main
https://huggingface.co/zgcr654321/8.ocr_text_detection_training/tree/main
https://huggingface.co/zgcr654321/9.ocr_text_recognition_training/tree/main
https://huggingface.co/zgcr654321/10.face_detection_training/tree/main
https://huggingface.co/zgcr654321/11.face_parsing_training/tree/main
https://huggingface.co/zgcr654321/12.human_parsing_training/tree/main
https://huggingface.co/zgcr654321/20.diffusion_model_training/tree/main
https://huggingface.co/zgcr654321/pretrained_models/tree/main

# Baidu-Netdisk
链接：https://pan.baidu.com/s/1yhEwaZhrb2NZRpJ5eEqHBw 
提取码：rgdo

Prepare datasets

CIFAR10

Make sure the folder architecture as follows:

CIFAR10
|
|-----batches.meta  unzip from cifar-10-python.tar.gz
|-----data_batch_1  unzip from cifar-10-python.tar.gz
|-----data_batch_2  unzip from cifar-10-python.tar.gz
|-----data_batch_3  unzip from cifar-10-python.tar.gz
|-----data_batch_4  unzip from cifar-10-python.tar.gz
|-----data_batch_5  unzip from cifar-10-python.tar.gz
|-----readme.html   unzip from cifar-10-python.tar.gz
|-----test_batch    unzip from cifar-10-python.tar.gz

CIFAR100

Make sure the folder architecture as follows:

CIFAR100
|
|-----train unzip from cifar-100-python.tar.gz
|-----test  unzip from cifar-100-python.tar.gz
|-----meta  unzip from cifar-100-python.tar.gz

ImageNet 1K(ILSVRC2012)

Make sure the folder architecture as follows:

ILSVRC2012
|
|-----train----1000 sub classes folders
|-----val------1000 sub classes folders
Please make sure the same class has same class folder name in train and val folders.

ImageNet 21K(Winter 2021 release)

Make sure the folder architecture as follows:

ImageNet21K
|
|-----train-----------10450 sub classes folders
|-----val-------------10450 sub classes folders
|-----small_classes---10450 sub classes folders
|-----imagenet21k_miil_tree.pth
Please make sure the same class has same class folder name in train and val folders.

ACCV2022

Make sure the folder architecture as follows:

ACCV2022
|
|-----train-------------5000 sub classes folders
|-----testa-------------60000 images
|-----accv2022_broken_list.json

VOC2007 and VOC2012

Make sure the folder architecture as follows:

VOCdataset
|                 |----Annotations
|                 |----ImageSets
|----VOC2007------|----JPEGImages
|                 |----SegmentationClass
|                 |----SegmentationObject
|        
|                 |----Annotations
|                 |----ImageSets
|----VOC2012------|----JPEGImages
|                 |----SegmentationClass
|                 |----SegmentationObject

COCO2017

Make sure the folder architecture as follows:

COCO2017
|                |----captions_train2017.json
|                |----captions_val2017.json
|--annotations---|----instances_train2017.json
|                |----instances_val2017.json
|                |----person_keypoints_train2017.json
|                |----person_keypoints_val2017.json
|                 
|                |----train2017
|----images------|----val2017

SAMACOCO

Make sure the folder architecture as follows:

SAMA-COCO
|                |----sama_coco_train.json
|                |----sama_coco_validation.json
|--annotations---|----train_labels.json
|                |----validation_labels.json
|                |----test_labels.json
|                |----image_info_test2017.json
|                |----image_info_test-dev2017.json
|                 
|                |----train
|----images------|----validation

Objects365(v2,2020)

Make sure the folder architecture as follows:

objects365_2020
|
|                |----zhiyuan_objv2_train.json
|--annotations---|----zhiyuan_objv2_val.json
|                |----sample_2020.json
|                 
|                |----train all train patch folders
|----images------|----val   all val patch folders
                 |----test  all test patch folders

ADE20K

Make sure the folder architecture as follows:

ADE20K
|                 |----training
|---images--------|----validation
|                 |----testing
|        
|                 |----training
|---annotations---|----validation

CelebA-HQ

Make sure the folder architecture as follows:

CelebA-HQ
|                 |----female
|---train---------|----male
|        
|                 |----female
|---val-----------|----male

FFHQ

Make sure the folder architecture as follows:

FFHQ
|
|---images
|---ffhq-dataset-v1.json
|---ffhq-dataset-v2.json

How to train or test a model

If you want to train or test a model,you need enter a training experiment folder directory,then run train.sh or test.sh.

For example,you can enter in folder classification_training/imagenet/resnet50.

If you want to restart train this model,please delete checkpoints and log folders first,then run train.sh:

CUDA_VISIBLE_DEVICES=0,1 python -m torch.distributed.run --nproc_per_node=2 --master_addr 127.0.1.0 --master_port 10000 ../../../tools/train_classification_model.py --work-dir ./

if you want to test this model,you need have a pretrained model first,modify trained_model_path in test_config.py,then run test.sh:

CUDA_VISIBLE_DEVICES=0 python -m torch.distributed.run --nproc_per_node=1 --master_addr 127.0.1.0 --master_port 10000 ../../../tools/test_classification_model.py --work-dir ./

CUDA_VISIBLE_DEVICES is used to specify the gpu ids for this training.Please make sure the number of nproc_per_node equal to the number of using gpu cards.Make sure master_addr/master_port are unique for each training.

Checkpoints/log folders are saved in your executing training/testing experiment folder directory.

Also, You can modify super parameters in train_config.py/test_config.py.

How to use gradio demo

cd to gradio_demo,we have:

classification demo
detection demo
semantic_segmentation demo
instance_segmentation demo
salient_object_detection demo
human_matting demo
text_detection demo
text_recognition demo
face_detection demo
face_parsing demo
human_parsing demo
point target segment_anything demo
circle target segment_anything demo

For example,you can run detection gradio demo(please prepare trained model weight first and modify model weight load path):

python gradio_detect_single_image.py

Reference

https://github.com/facebookresearch/segment-anything
https://github.com/facebookresearch/sam2

Citation

If you find my work useful in your research, please consider citing:

@inproceedings{zgcr,
 title={SimpleAICV-pytorch-training-examples},
 author={zgcr},
 year={2020-2024}
}