Overview

Name	Dataset	Description	Scene	Download	Remark
coco_segmentation	COCO	See here	Horizontal FL	Click Here	(under testing)
oxfordiiitpet_segmentation	OxfordIIITPet	See here	Horizontal FL	Click Here
sbdataset_segmentation	SBDataset	See here	Horizontal FL	Click Here
cityspaces_segmentation	Cityspaces	See here	Horizontal FL	Click Here
camvid_segmentation	CamVID	See here	Horizontal FL	Click Here
ade20k_segmentation	ADE20k	See here	Horizontal FL	Click Here
voc_segmentation	PASCAL_VOC	See here	Horizontal FL	Click Here

Details

coco_segmentation

coco

model

Model Name	Non-Fed Performance	NumPara	Implementation
FCN_ResNet50	-		-
UNet	-		-

supported partitioner

Name	IsDefault	Comments
IIDPartitioner	yes
DiversityPartitioner		Partitioning according to label diversity
DirichletPartitioner		Partitioning according to dir. distribution of labels

oxfordiiitpet_segmentation

coco

model

Model Name	Non-Fed Performance	NumPara	Implementation
FCN_ResNet50	-		-

supported partitioner

Name	IsDefault	Comments
IIDPartitioner	yes
DiversityPartitioner		Partitioning according to label diversity
DirichletPartitioner		Partitioning according to dir. distribution of labels

sbdataset_segmentation

coco

model

Model Name	Non-Fed Performance	NumPara	Implementation
FCN_ResNet50	-		-
UNet	-		-

supported partitioner

Name	IsDefault	Comments
IIDPartitioner	yes
DiversityPartitioner		Partitioning according to label diversity
DirichletPartitioner		Partitioning according to dir. distribution of labels

cityspaces_segmentation

Usage

To use this benchmark, you need to manually download the raw data into the dictionary 'cityspaces_segmentation/cityspaces/'. The necessary file contains leftImg8bit_trainvaltest.zip (11GB) and gtFine_trainvaltest.zip (241MB).

model

Model Name	Non-Fed Performance	NumPara	Implementation
FCN_ResNet50	-		-

supported partitioner

Name	IsDefault	Comments
IIDPartitioner	yes

camvid_segmentation

Usage

To use this benchmark, you need to manually download the raw data into the dictionary 'camvid_segmentation/CamVid/' from Kaggle. The downloaded .zip file should also be manually into 'camvid_segmentation/CamVid/'. The architecture of the benchmark should be like:

├─ camvid_segmentation
│  ├─ CamVid            //classification on mnist dataset
│  │  ├─ train       
│  |  │  ├─ xxx.png           // horizontal fedtask
│  |  │  ...  
│  │  ├─ train_labels   
│  |  │  ├─ xxx.png    
│  |  │  ...  
│  │  ├─ val
│  │  ├─ val_labels 
│  │  ├─ test
│  │  ├─ test_labels 
│  │  └─ class_dict.csv
│  |     
│  ├─ model
│  ├─ config.py
│  ├─ core.py
│  └─ __init__.py

model

Model Name	Non-Fed Performance	NumPara	Implementation
FCN_ResNet50	-		-

supported partitioner

Name	IsDefault	Comments
IIDPartitioner	yes

ade20k_segmentation

ADE20K

model

Model Name	Non-Fed Performance	NumPara	Implementation
FCN_ResNet50	-		-

supported partitioner

Name	IsDefault	Comments
IIDPartitioner	yes
DiversityPartitioner		Partitioning according to label diversity
DirichletPartitioner		Partitioning according to dir. distribution of labels

voc_segmentation

VOC

model

Model Name	Non-Fed Performance	NumPara	Implementation
FCN_ResNet50	-		-

supported partitioner

Name	IsDefault	Comments
IIDPartitioner	yes