search

tiancity-NJU / da-faster-rcnn-PyTorch

An unofficial implementation of 'Domain Adaptive Faster R-CNN for Object Detection in the Wild ’
Other
136 stars 37 forks source link

number of test images #18

Open kailigo opened 5 years ago

kailigo commented 5 years ago

Hi, the provided dataset has 347 images for testing, while the paper says that there are 500 images for testing. I wonder why there is such a difference. Thanks.

kailigo commented 5 years ago

To make it clear.

The foggy_cityscape and cityscape datasets should have identical structure, i.e., same number of train/val images. However, I noticed that in your provided dataset, the foggy_cityscape dataset (test_t.txt) has 500 images as val set, while cityscape has only 374 images (test_s.txt). Could you tell how this difference happens. Thanks.

kailigo commented 5 years ago

followup.

I also noticed that the test images all appears in the training images, ie. test_s is a subset of train_s. could you help take a look of the dataset and upload a new dataset version. Appreciated.

stormchasingg commented 5 years ago

Images
train_s: leftImg8bit/train
train_t: leftImg8bit_foggy/train
test_t: leftImg8bit_foggy/val 3 classes, 'foggy_beta_0.02', totally 500 images Annotations
gtFine/train+val
i download dataset from cityscapes official web

Fly-dream12 commented 3 years ago

Could you share the foggy dataset and the annotations, i can not downloaded it completely. Thanks @stormchasingg