Performance on Cityscapes to Cross-City
BinhuiXie opened this issue · comments
Hi, thank you for your great work. I'm trying to reproduce the results of Cityscapes to Cross-City, but the performance is not as good as yours. Here are some questions:
In your paper (Table 1), for Cityscapes to Rio task, the mIoUs of Source DeepLab-v2 and AdaptSegNet are 48.2
and 51.6
respectively. But I got mIoU of 44.9
for Source DeepLab-v2 and mIoU of 47.0
for AdaptSegNet. Therfore, there is performance degradation for FADA and FADA w/ self distillation.
The config file of Source DeepLab-v2
:
MODEL:
NAME: "deeplab_resnet101"
WEIGHTS: 'https://download.pytorch.org/models/resnet101-5d3b4d8f.pth'
FREEZE_BN: True
NUM_CLASSES: 19
DATASETS:
SOURCE_TRAIN: "cityscapes_train"
TEST: "crosscity_Rio_test"
INPUT:
SOURCE_INPUT_SIZE_TRAIN: (1024, 512)
TARGET_INPUT_SIZE_TRAIN: (1024, 512)
INPUT_SIZE_TEST: (2048, 1024)
SOLVER:
BASE_LR: 5e-4
MAX_ITER: 31250
STOP_ITER: 20000
BATCH_SIZE: 8
BATCH_SIZE_VAL: 1
And the config file of AdaptSegNet
:
MODEL:
NAME: "deeplab_resnet101"
WEIGHTS: "https://download.pytorch.org/models/resnet101-5d3b4d8f.pth"
FREEZE_BN: True
NUM_CLASSES: 19
DATASETS:
SOURCE_TRAIN: "cityscapes_train"
TARGET_TRAIN: "crosscity_Rio_train"
TEST: "crosscity_Rio_test"
INPUT:
SOURCE_INPUT_SIZE_TRAIN: (1024, 512)
TARGET_INPUT_SIZE_TRAIN: (1024, 512)
INPUT_SIZE_TEST: (2048, 1024)
BRIGHTNESS: 0.5
CONTRAST: 0.5
SATURATION: 0.5
HUE: 0.2
SOLVER:
BASE_LR: 2.5e-4
BASE_LR_D: 1e-4
MAX_ITER: 62500
STOP_ITER: 40000
BATCH_SIZE: 8
If possible, could you provide the config files and data loader for reproducing the results?
Thank you anyway.
Hi, thanks for your interest!
The result of baseline and AdaptSegNet on Cityscapes->Cross-City is copied from AdaptSegNet paper. I also find it difficult to reproduce the results of AdaptSegNet with this codebase. As the author of AdaptSegNet doesn't release codes for Cityscapes->Cross-City, it would be highly possible that my implementation is inconsistent with theirs. I would suggest you implement their methods with their codebase and their provided pre-trained weights to reproduce their results.
For your config file, I think the problem might be your INPUT_SIZE_TEST. I would suggest you use same input size for training and testing for same dataset, as the DeeplabV2 would be sensitive to the scales of objects.
I would contact other authors of this paper to see whether configs and pre-trained weights are kept in the server. I will come back to you in 1-2 weeks.
I'm grateful to you for your replay.
Hi, the cross_city related code could be found in this file: cross_city_codes.tar.gz. It was implemented on a previous version of this project. Unfortunately recently I don't have time to test these code snippets with this repo. Hope it could give you some hints. Best wishes!