• My ZhiHu column
• Introduction
• Environments
• Download datasets and my pretrained models
• Prepare datasets
• Train and test model
• Contrastive learning training results
  • ILSVRC2012(ImageNet) pretrained results
  • ILSVRC2012(ImageNet) finetune results
• Masked image modeling training results
  • ILSVRC2012(ImageNet) pretrained results
  • ILSVRC2012(ImageNet) finetune results
• Classification training results
  • ILSVRC2012(ImageNet) training results
  • CIFAR100 training results
• Detection training results
  • COCO2017 training results
  • VOC2007 and VOC2012 training results
• Semantic Segmentation training results
  • ADE20K training results
• Distillation training results
  • ImageNet training results
• Citation

My ZhiHu column

https://www.zhihu.com/column/c_1249719688055193600

Introduction

This repository provides simple training and testing examples for image classification, object detection, semantic segmentation, knowledge distillation, contrastive learning, masked image modeling training.

contrastive learning

DINO(for ResNet50)

masked image modeling training

MAE(for ViT_Base_Patch16)

image classification:

ResNet
DarkNet
RepVGG
RegNetX

object detection:

RetinaNet
FCOS
CenterNet
TTFNet

semantic segmentation:

DeepLabv3+
Segmenter

knowledge distillation:

KD loss
DKD loss
DML loss

Environments

This repository only support one server one gpu card/one server multi gpu cards.

environments:

Ubuntu 20.04.3 LTS,30 core AMD EPYC 7543 32-Core Processor, 2*RTX A5000, Python Version:3.8, CUDA Version:11.3

Please make sure your Python version>=3.7.

Use pip or conda to install those Packages:

torch==1.10.0
torchvision==0.11.1
torchaudio==0.10.0
onnx==1.11.0
onnx-simplifier==0.3.6
numpy
Cython
pycocotools
opencv-python
tqdm
thop==0.0.31.post2005241907
yapf
apex

How to install apex?

apex needs to be installed separately.First,download apex：

git clone https://github.com/NVIDIA/apex

For torch1.10,modify apex/apex/amp/utils.py:

if cached_x.grad_fn.next_functions[1][0].variable is not x:

to

if cached_x.grad_fn.next_functions[0][0].variable is not x:

Then use the following orders to install apex:

cd apex
pip install -v --no-cache-dir ./

Using apex to train can reduce video memory usage by 25%-30%, but the training speed will be slower, the trained model has the same performance as not using apex.

Download datasets and my pretrained models

You can download all datasets、all my pretrained models and all my experiments records/checkpoints from Baidu-Netdisk:

# all datasets
链接：https://pan.baidu.com/s/16fAq74UMe6M1FVO_IDbqgA 
提取码：vtm8
# all my pretrained models
链接：https://pan.baidu.com/s/1j7jm3I-UpMH7lJUOmJxhqg 
提取码：okm8
# all my experiments records/checkpoints
链接：https://pan.baidu.com/s/17yT6m98lXi9gXYgIK5afyw 
提取码：ofc0

Prepare datasets

If you want to reproduce my imagenet pretrained models,you need download ILSVRC2012 dataset,and 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.

If you want to reproduce my cifar100 pretrained models,you need download cifar100 dataset,and 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

If you want to reproduce my COCO2017 pretrained models,you need download COCO2017 dataset,and 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

If you want to reproduce my VOC2007/VOC2012 pretrained models,you need download VOC2007+VOC2012 dataset,and make sure the folder architecture as follows:

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

If you want to reproduce my ADE20K pretrained models,you need download ADE20K dataset,and make sure the folder architecture as follows:

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

Train and test model

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

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

If you want to train this model from scratch,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 ./

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 gpu cards.Make sure master_addr/master_port are unique for each training.

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

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

If you see the following log, and train.sh/test.sh keeps running, then train.sh/test.sh is running correctly.

All checkpoints/log are saved in training/testing folder directory.

Warning:  apex was installed without --cpp_ext.  Falling back to Python flatten and unflatten.
Selected optimization level O1:  Insert automatic casts around Pytorch functions and Tensor methods.

Defaults for this optimization level are:
enabled                : True
opt_level              : O1
cast_model_type        : None
patch_torch_functions  : True
keep_batchnorm_fp32    : None
master_weights         : None
loss_scale             : dynamic
Processing user overrides (additional kwargs that are not None)...
After processing overrides, optimization options are:
enabled                : True
opt_level              : O1
cast_model_type        : None
patch_torch_functions  : True
keep_batchnorm_fp32    : None
master_weights         : None
loss_scale             : dynamic
Warning:  multi_tensor_applier fused unscale kernel is unavailable, possibly because apex was installed without --cuda_ext --cpp_ext. Using Python fallback.
Original ImportError was: ModuleNotFoundError("No module named 'amp_C'")
Warning:  apex was installed without --cpp_ext.  Falling back to Python flatten and unflatten.

Contrastive learning training results

DINO:Emerging Properties in Self-Supervised Vision Transformers

Paper:https://arxiv.org/abs/2104.14294

ILSVRC2012(ImageNet) pretrained results

Network	input size	gpu num	batch	epochs	Loss
ResNet50_dino_pretrained_epoch100	224x224	2 RTX A5000	128	100	2.4567

ILSVRC2012(ImageNet) finetune results

Network	macs	params	input size	gpu num	batch	epochs	Top-1
ResNet50_finetune_epoch100_dino_pretrained_epoch_100	4.112G	25.557M	224x224	2 RTX A5000	256	100	76.858

Masked image modeling training results

MAE:Masked Autoencoders Are Scalable Vision Learners

Paper:https://arxiv.org/abs/2111.06377

VIT:An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

Paper:https://arxiv.org/abs/2010.11929

ILSVRC2012(ImageNet) pretrained results

Network	input size	gpu num	batch	epochs	Loss
ViT_Base_Patch16_mae_pretrained_epoch100	224x224	2 RTX A5000	256x4	100	0.3986
ViT_Base_Patch16_mae_pretrained_epoch400	224x224	2 RTX A5000	256x4	400	0.3879

ILSVRC2012(ImageNet) finetune results

Network	macs	params	input size	gpu num	batch	epochs	Top-1
ViT_Base_Patch16_finetune_epoch100_mae_pretrained_epoch_100	16.849G	86.377M	224x224	2 RTX A5000	256x4	100	82.182
ViT_Base_Patch16_finetune_epoch100_mae_pretrained_epoch_400	16.849G	86.377M	224x224	2 RTX A5000	256x4	100	83.130

Classification training results

ResNet

Paper:https://arxiv.org/abs/1512.03385

DarkNet

Paper:https://arxiv.org/abs/1804.02767?e05802c1_page=1

RepVGG

Paper:https://arxiv.org/abs/2101.03697

RegNet

Paper:https://arxiv.org/abs/2003.13678

ILSVRC2012(ImageNet) training results

Network	macs	params	input size	gpu num	batch	epochs	Top-1
ResNet18	1.819G	11.690M	224x224	2 RTX A5000	256	100	70.712
ResNet34half	949.323M	5.585M	224x224	2 RTX A5000	256	100	67.752
ResNet34	3.671G	21.798M	224x224	2 RTX A5000	256	100	73.752
ResNet50half	1.063G	6.918M	224x224	2 RTX A5000	256	100	72.902
ResNet50	4.112G	25.557M	224x224	2 RTX A5000	256	100	76.264
ResNet101	7.834G	44.549M	224x224	2 RTX A5000	256	100	77.322
ResNet152	11.559G	60.193M	224x224	2 RTX A5000	256	100	78.006
DarkNetTiny	412.537M	2.087M	256x256	2 RTX A5000	256	100	57.602
DarkNet19	3.663G	20.842M	256x256	2 RTX A5000	256	100	74.028
DarkNet53	9.322G	41.610M	256x256	2 RTX A5000	256	100	76.602
RepVGG_A0_deploy	1.362G	8.309M	224x224	2 RTX A5000	256	120	72.156
RepVGG_A1_deploy	2.364G	12.790M	224x224	2 RTX A5000	256	120	74.056
RepVGG_A2_deploy	5.117G	25.500M	224x224	2 RTX A5000	256	120	76.022
RepVGG_B0_deploy	3.058G	14.339M	224x224	2 RTX A5000	256	120	74.750
RepVGG_B1_deploy	11.816G	51.829M	224x224	2 RTX A5000	256	120	77.834
RepVGG_B2_deploy	18.377G	80.315M	224x224	2 RTX A5000	256	120	78.226
RegNetX_400MF	410.266M	5.158M	224x224	2 RTX A5000	256	100	72.364
RegNetX_600MF	616.813M	6.196M	224x224	2 RTX A5000	256	100	73.598
RegNetX_800MF	820.324M	7.260M	224x224	2 RTX A5000	256	100	74.444
RegNetX_1_6GF	1.635G	9.190M	224x224	2 RTX A5000	256	100	76.580
RegNetX_3_2GF	3.222G	15.297M	224x224	2 RTX A5000	256	100	77.512
RegNetX_4_0GF	4.013G	22.118M	224x224	2 RTX A5000	256	100	77.722

You can find more model training details in classification_training/imagenet/.

CIFAR100 training results

Network	macs	params	input size	gpu num	batch	epochs	Top-1
ResNet18Cifar	557.935M	11.220M	32x32	1 RTX A5000	128	200	76.730
ResNet34halfCifar	292.370M	5.350M	32x32	1 RTX A5000	128	200	75.730
ResNet34Cifar	1.164G	21.328M	32x32	1 RTX A5000	128	200	77.850
ResNet50halfCifar	331.879M	5.991M	32x32	1 RTX A5000	128	200	75.880
ResNet50Cifar	1.312G	23.705M	32x32	1 RTX A5000	128	200	75.890
ResNet101Cifar	2.531G	42.697M	32x32	1 RTX A5000	128	200	79.710
ResNet152Cifar	3.751G	58.341M	32x32	1 RTX A5000	128	200	77.150

You can find more model training details in classification_training/cifar100/.

Detection training results

RetinaNet

Paper:https://arxiv.org/abs/1708.02002

FCOS

Paper:https://arxiv.org/abs/1904.01355

CenterNet

Paper:https://arxiv.org/abs/1904.07850

TTFNet

Paper:https://arxiv.org/abs/1909.00700

YOLOX

Paper:https://arxiv.org/abs/2107.08430

How to use yolov3 anchor clustering method to generate a set of custom anchors for your own dataset?

I provide a script in simpleAICV/detection/yolov3_anchor_cluster.py,and I give two examples for generate anchors on COCO2017 and VOC2007+2012 datasets.If you want to generate anchors for your dataset,just modify the part of input code,get width and height of all annotaion boxes,then use the script to compute anchors.The anchors size will change with different datasets or different input resizes.

COCO2017 training results

Trained on COCO2017_train dataset, tested on COCO2017_val dataset.

mAP is IoU=0.5:0.95,area=all,maxDets=100,mAP(COCOeval,stats[0]).

Network	resize-style	input size	macs	params	gpu num	batch	epochs	mAP
ResNet50-RetinaNet	RetinaStyle-400	400x667	63.093G	37.969M	2 RTX A5000	32	13	31.939
ResNet50-RetinaNet	RetinaStyle-800	800x1333	250.069G	37.969M	2 RTX A5000	8	13	35.082
ResNet50-RetinaNet	YoloStyle-640	640x640	95.558G	37.969M	2 RTX A5000	32	13	33.475
ResNet101-RetinaNet	RetinaStyle-800	800x1333	329.836G	56.961M	2 RTX A5000	8	13	36.406
ResNet50-FCOS	RetinaStyle-400	400x667	54.066G	32.291M	2 RTX A5000	32	13	34.671
ResNet50-FCOS	RetinaStyle-800	800x1333	214.406G	32.291M	2 RTX A5000	8	13	37.850
ResNet50-FCOS	YoloStyle-640	640x640	81.943G	32.291M	2 RTX A5000	32	13	35.629
ResNet101-FCOS	RetinaStyle-800	800x1333	294.173G	51.283M	2 RTX A5000	8	13	39.547
vit_base_patch16_fcos	YoloStyle-640	640x640	184.171G	91.506M	2 RTX A5000	8	13	39.948
ResNet18DCN-CenterNet	YoloStyle-512	512x512	14.854G	12.889M	2 RTX A5000	64	140	27.947
ResNet18DCN-TTFNet-3x	YoloStyle-512	512x512	16.063G	13.737M	2 RTX A5000	64	39	27.847

You can find more model training details in detection_training/coco/.

VOC2007 and VOC2012 training results

Trained on VOC2007 trainval dataset + VOC2012 trainval dataset, tested on VOC2007 test dataset.

mAP is IoU=0.50,area=all,maxDets=100,mAP.

Network	resize-style	input size	macs	params	gpu num	batch	epochs	mAP
ResNet50-RetinaNet	YoloStyle-640	640x640	84.947G	36.724M	2 RTX A5000	32	13	80.693
ResNet50-FCOS	YoloStyle-640	640x640	80.764G	32.153M	2 RTX A5000	32	13	79.960

You can find more model training details in detection_training/voc/.

Semantic Segmentation training results

DeepLabv3+

Paper:https://arxiv.org/abs/1802.02611

Segmenter

Paper:https://arxiv.org/abs/2105.05633

ADE20K training results

Network	input size	macs	params	gpu num	batch	epochs	miou
ResNet50-DeepLabv3+	512x512	25.548G	26.738M	2 RTX A5000	8	128	34.058
vit_base_patch16_Segmenter_mae_pretrain_epoch400	512x512	104.886G	100.368M	2 RTX A5000	8	128	40.652
vit_base_patch16_Segmenter_cls_pretrain_83.130	512x512	104.886G	100.368M	2 RTX A5000	8	128	42.058

Distillation training results

KD loss

Paper:https://arxiv.org/abs/1503.02531

DKD loss

Paper:https://arxiv.org/abs/2203.08679

DML loss

Paper:https://arxiv.org/abs/1706.00384

ImageNet training results

Teacher Network	Student Network	method	Freeze Teacher	input size	gpu num	batch	epochs	Teacher Top-1	Student Top-1
ResNet34	ResNet18	CE+KD	True	224x224	2 RTX A5000	256	100	/	71.984
ResNet34	ResNet18	CE+DKD	True	224x224	2 RTX A5000	256	100	/	72.110
ResNet34	ResNet18	CE+DML	False	224x224	2 RTX A5000	256	100	74.674	72.064
ResNet152	ResNet50	CE+KD	True	224x224	2 RTX A5000	256	100	/	76.476
ResNet152	ResNet50	CE+DKD	True	224x224	2 RTX A5000	256	100	/	77.616
ResNet152	ResNet50	CE+DML	False	224x224	2 RTX A5000	256	100	79.148	77.622

You can find more model training details in distillation_training/imagenet/.

Citation

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

@inproceedings{zgcr,
 title={SimpleAICV-ImageNet-CIFAR-COCO-VOC-training},
 author={zgcr},
 year={2022}
}