MEDA: Manifold Embedded Distribution Alignment

This directory contains the code for paper Visual Domain Adaptation with Manifold Embedded Distribution Alignment published at ACM Multimedia conference (ACM MM) 2018 as an Oral presentation. This paper is also lucky to be ranked as Top 10 papers.

Usage

The original code is written using Matlab R2017a. I think all versions after 2015 can run the code.

For Python users, I add a MEDA.py implementation. The Python version will need to import GFK module (can be found here). However, this Python version is only for reference since the graph Laplacian (as exactly in Matlab) is not implemented.

Demo

I offer a basic demo to run on the Office+Caltech10 datasets. Download the datasets here and put the data (mat files) into the data folder.

Run demo_office_caltech_surf.m.

Results

MEDA achieved state-of-the-art performances compared to a lot of traditional and deep methods as of 2018. The testing datasets are most popular domain adaptation and transfer learning datasets: Office+Caltech10, Office-31, USPS+MNIST, ImageNet+VOC2007.

The following results are from the original paper and its supplementary file.

Office-31 dataset

Using ResNet-50 features (compare with the latest deep methods with ResNet-50 as backbone). It seems MEDA is the only traditional method that can challenge these heavy deep adversarial methods.

Download Office-31 ResNet-50 features

	Method	A - W	D - W	W-D	A - D	D - A	W-A	AVG
cvpr16	ResNet-50	68.4	96.7	99.3	68.9	62.5	60.7	76.1
icml15	DAN	80.5	97.1	99.6	78.6	63.6	62.8	80.4
nips16	RTN	84.5	96.8	99.4	77.5	66.2	64.8	81.6
icml15	DANN	82.0	96.9	99.1	79.7	68.2	67.4	82.2
cvpr17	ADDA	86.2	96.2	98.4	77.8	69.5	68.9	82.9
icml17	JAN	85.4	97.4	99.8	84.7	68.6	70.0	84.3
cvpr17	GTA	89.5	97.9	99.8	87.7	72.8	71.4	86.5
nips18	CDAN-RM	93.0	98.4	100.0	89.2	70.2	67.4	86.4
nips18	CDAN-M	93.1	98.6	100.0	92.9	71.0	69.3	87.5
cvpr18	CAN	81.5	63.4	85.5	65.9	99.7	98.2	82.4
aaai19	JDDA	82.6	95.2	99.7	79.8	57.4	66.7	80.2
aaai18	MADA	90.1	97.4	99.6	87.8	70.3	66.4	85.2
ACMMM18	MEDA	86.2	97.2	99.4	85.3	72.4	74.0	85.7

Office-Home

Using ResNet-50 features (compare with the latest deep methods with ResNet-50 as backbone). Again, it seems that MEDA achieves the best performance.

Download Office-Home ResNet-50 pretrained features

	Method	Ar-Cl	Ar-Pr	Ar-Rw	Cl-Ar	Cl-Pr	Cl-Rw	Pr-Ar	Pr-Cl	Pr-Rw	Rw-Ar	Rw-Cl	Rw-Pr	Avg
	AlexNet	26.4	32.6	41.3	22.1	41.7	42.1	20.5	20.3	51.1	31.0	27.9	54.9	34.3
icml15	DAN	31.7	43.2	55.1	33.8	48.6	50.8	30.1	35.1	57.7	44.6	39.3	63.7	44.5
icml15	DANN	36.4	45.2	54.7	35.2	51.8	55.1	31.6	39.7	59.3	45.7	46.4	65.9	47.3
icml17	JAN	35.5	46.1	57.7	36.4	53.3	54.5	33.4	40.3	60.1	45.9	47.4	67.9	48.2
nips18	CDAN-RM	36.2	47.3	58.6	37.3	54.4	58.3	33.2	43.9	62.1	48.2	48.1	70.7	49.9
nips18	CDAN-M	38.1	50.3	60.3	39.7	56.4	57.8	35.5	43.1	63.2	48.4	48.5	71.1	51.0
cvpr16	ResNet-50	34.9	50.0	58.0	37.4	41.9	46.2	38.5	31.2	60.4	53.9	41.2	59.9	46.1
icml15	DAN	43.6	57.0	67.9	45.8	56.5	60.4	44.0	43.6	67.7	63.1	51.5	74.3	56.3
icml15	DANN	45.6	59.3	70.1	47.0	58.5	60.9	46.1	43.7	68.5	63.2	51.8	76.8	57.6
icml17	JAN	45.9	61.2	68.9	50.4	59.7	61.0	45.8	43.4	70.3	63.9	52.4	76.8	58.3
nips18	CDAN-RM	49.2	64.8	72.9	53.8	62.4	62.9	49.8	48.8	71.5	65.8	56.4	79.2	61.5
nips18	CDAN-M	50.6	65.9	73.4	55.7	62.7	64.2	51.8	49.1	74.5	68.2	56.9	80.7	62.8
ACMMM18	MEDA	54.6	75.2	77.0	56.5	72.8	72.3	59.0	51.9	78.2	67.7	57.2	81.8	67.0

Image-CLEF DA

using ResNet-50 features (compare with the latest deep methods with ResNet-50 as backbone). Again, it seems that MEDA achieves the best performance.

Download Image-CLEF ResNet-50 pretrained features

Method	I-P	P-I	I-C	C-I	C-P	P-C	Avg
AlexNet	66.2	70.0	84.3	71.3	59.3	84.5	73.9
DAN	67.3	80.5	87.7	76.0	61.6	88.4	76.9
DANN	66.5	81.8	89.0	79.8	63.5	88.7	78.2
JAN	67.2	82.8	91.3	80.0	63.5	91.0	79.3
CDAN-RM	67.0	84.8	92.4	81.3	64.7	91.6	80.3
CDAN-M	67.7	83.3	91.8	81.5	63.0	91.5	79.8
ResNet-50	74.8	83.9	91.5	78.0	65.5	91.2	80.7
DAN	74.5	82.2	92.8	86.3	69.2	89.8	82.5
DANN	75.0	86.0	96.2	87.0	74.3	91.5	85.0
RTN	75.6	86.8	95.3	86.9	72.7	92.2	84.9
JAN	76.8	88.0	94.7	89.5	74.2	91.7	85.8
MADA	75.0	87.9	96.0	88.8	75.2	92.2	85.8
CDAN-RM	77.2	88.3	98.3	90.7	76.7	94.0	87.5
CDAN-M	78.3	91.2	96.7	91.2	77.2	93.7	88.1
CAN	78.2	87.5	94.2	89.5	75.8	89.2	85.7
iCAN	79.5	89.7	94.7	89.9	78.5	92.0	87.4
MEDA	79.7	92.5	95.7	92.2	78.5	95.5	89.0

Office-31 dataset using DECAF features (compare with deep methods with AlexNet):

Office+Caltech 10 datasets and MNIST+USPS and ImageNet+VOC:

Reference

If you use this code, please cite it as:

Jindong Wang, Wenjie Feng, Yiqiang Chen, Han Yu, Meiyu Huang, Philip S. Yu. Visual Domain Adaptation with Manifold Embedded Distribution Alignment. ACM Multimedia conference 2018.

Or in bibtex style:

@inproceedings{wang2018visual,
    title={Visual Domain Adaptation with Manifold Embedded Distribution Alignment},
    author={Wang, Jindong and Feng, Wenjie and Chen, Yiqiang and Yu, Han and Huang, Meiyu and Yu, Philip S},
    booktitle={ACM Multimedia Conference (ACM MM)},
    year={2018}
}

wenchieh/MEDA