RGBT 对比算法 实验设置

2021-TIP-MIDD

backbone is VGG16(效果好) and ResNet50

VT821-VT1000-VT5000

训练集:VT5000-TR的训练集2500做训练集
测试集:VT5000-TE,VT821,VT1000

MTMR(RGBT)(t传统方法)
M3S-NIR(RGBT)(t)
SGDL(RGBT)(t)
ADF(RGBT)
MIDD(RGBT)

DMRA(RGBD)
S2MA(RGBD)

PFA(RGB-早期融合)
R3Net(RGB)
BASNet(RGB)
PoolNet(RGB)
CPD(RGB)
EGNet(RGB)

------------------------------------------------------------------------------------------

HAINet

在VT1000中训练
在VT821种测试

We apply our HAINet to RGB-T SOD, and provide results (code: s82s) of our HAINet on VT821 dataset trained with VT1000 dataset.

对比算法
MTMR(RGB-T)(t)
M2S-NIR(RGB-T)(t)

SGDL(RGB)
ADF
MIED(RGBT)

-------------------------------------------------------------------------------------------

2021-TIP-FML
backbone is VGG16

训练集合: we randomly select 410 RGB-thermal image pairs from the RGB-thermal dataset and 200 RGB-thermal image pairs from the selected Grayscale-thermal dataset (i.e., the car and pedes trian sets) as the training set.

测试集合: The rest of RGB-thermal image pairs in RGB-thermal dataset and Grayscale-thermal dataset are used as the testing set. '

辅助训练：Then, 830 samples are randomly selected from the training set of the MSRA-B as an auxiliary set to train the RGB/thermal branch of the proposed model.

三个训练集
RGB-thermal dataset（821RGBT）Grayscale-thermal dataset，MSRA-B dataset（RGB）

R3Net(RGB-RGBT)
PoolNet(RGB-RGBT)
CPD(RGB-RGBT)

AFNet(RGBD-RGBT)
PDNet(RGBD-RGBT)
TSAA(RGBD-RGBT)
SSRC(RGBD-RGBT)

NRCMC(RGBT)
MFSR(RGBT)
CGL(RGBT)