Can I use the provided source only model to start training adaptation? (https://drive.google.com/file/d/1-52RggreImwr_BVcGzm41j0mchxclwwu/view) So that I can skip the first step where a large memory GPU is required to train the model.

Hi
  I am a little confused about these two models, the ImageNet-pretrained Model and the Source-only Model. Why does the Source-only Model appear in this ccm_config.yml file？CCM and Source only, aren't they two different methods ?

for the below three rows:
if self.config.neptune:
neptune.init(project_qualified_name="solacex/segmentation-DA")
neptune.create_experiment(params=self.config, name=self.config["note"])

get the below rseult:
neptune.exceptions.MissingApiToken: Missing API token. Use "NEPTUNE_API_TOKEN" environment variable or pass it as an argument

it's so strange for me, can you tell me how to tackle this problem?

Hi there,

Thanks for your code!

I ran the code and only got 48.8 mIoU which is 1 mIoU inferior to the reported one in the paper.

I found that the code randomly selects part of the source dataset at each round before conducting image selection. And 'pool_prop' in ccm_config.yml determines this proportion, namely, 20% in default. Then around 1k images are selected from the remained 20% images.

Would this random selection process affect the performance? And is it enough to use only 1k source images for the training during each epoch? What is the best set of these hyper-parameters?

hi, I 'm a beginner.
I noticed that in https://github.com/2669531054/CCM/blob/2fcd9cb3b1d2070dd403613f00ca55ef872b3104/trainer/ccm_trainer.py#L203
, you use the label of the train set of target domain to calculate the accuracy of the pseudo label.
I wonder whether this accuracy influence the training process. And if I do not have the label of the target domain, can I skip this step?

FileNotFoundError: [Errno 2] No such file or directory: 'config/final_config.yml'

I found there are only ccm_config.yml and so_config.yml, so can you provide the final_config.yml?

Thanks so much!

Thanks for your work!
Could you provide the pre-trained model "SYNTHIA-Source-only"

Hi, thanks for your great jobs!
When I try to train a model, there was an error like that:

Traceback (most recent call last):
File "so_run.py", line 51, in
main()
File "so_run.py", line 43, in main
trainer.train()
File "/home/CCM/trainer/source_only_trainer.py", line 58, in train
self.optim.step()
File /home/anaconda3/envs/torch1.9/lib/python3.8/site-packages/torch/optim/optimizer.py", line 88, in wrapper
return func(*args, **kwargs)
File "/home/anaconda3/envs/torch1.9/lib/python3.8/site-packages/torch/autograd/grad_mode.py", line 28, in decorate_context
return func(*args, **kwargs)
File "/home/anaconda3/envs/torch1.9/lib/python3.8/site-packages/torch/optim/sgd.py", line 110, in step
F.sgd(params_with_grad,
File "/home/anaconda3/envs/torch1.9/lib/python3.8/site-packages/torch/optim/functional.py", line 180, in sgd
param.add(d_p, alpha=-lr)
RuntimeError: For non-complex input tensors, argument alpha must not be a complex number.

How should I fix it? Thank you.
And my config used to train is:

note: 'train'

configs of data

model: 'deeplab'
train: True
multigpu: False
fixbn: True
fix_seed: True

Optimizaers

learning_rate: 7.5e-5
num_steps: 5000
epochs: 2
weight_decay: 0.0005
momentum: 0.9
power: 0.9
round: 6

Logging

print_freq: 1
save_freq: 2000
tensorboard: False
neptune: False
screen: True
val: False
val_freq: 300

Dataset

source: 'gta5'
target: 'cityscapes'
worker: 0
batch_size: 2

#Transforms
input_src: 720
input_tgt: 720
crop_src: 600
crop_tgt: 600
mirror: True
scale_min: 0.5
scale_max: 1.5
rec: False

Model hypers

init_weight: './pretrained/DeepLab_resnet_pretrained_init-f81d91e8.pth'
restore_from: None

snapshot: './Data/snapshot/'
result: './miou_result/'
log: './log/'
plabel: './plabel'
gta5: {
data_dir: '/home/data/datasets/GTA5/',
data_list: './dataset/list/gta5_list.txt',
input_size: [1280, 720]
}
synthia: {
data_dir: '/home/guangrui/data/synthia/',
data_list: './dataset/list/synthia_list.txt',
input_size: [1280, 760]
}
cityscapes: {
data_dir: '/home/data/datasets/Cityscapes',
data_list: './dataset/list/cityscapes_train.txt',
input_size: [1024, 512]
}

Hi, I'm trying to run your code from source only train but there is no "list" folder in the dataset directory.

gta5: {
        data_dir: '/media/NAS/nas_187/datasets/gta/',
        data_list: './dataset/list/gta5_list.txt',
        input_size: [1280, 720]
        }
synthia: {
        data_dir: '/home/guangrui/data/synthia/',
        data_list: './dataset/list/synthia_list.txt',
        input_size: [1280, 760]
        }
cityscapes: {
        data_dir: '/media/NAS/nas_187/datasets/synthia',
        data_list: './dataset/list/cityscapes_train.txt',
        input_size: [1024, 512]
        }

Could you please add the missing files?

Also, there is a line trying to add nonexistent files in model/__init__.py

I'm a beginner.
I am not sure why the following parameters are set. Why is it 1191 or 0.9 ?
Thanks!

A small question!

Your ccm_trainer.py

159 index = int(cls_total * prop)
160 cls_thres = cls_prob[-index]
161 cls_thres2 = cls_prob[index]
162 thres[k] = cls_thres

For what consideration, you set your thres be the value of the last ten percent of the probability value?

last bacause of (-index), ten percent because of (prop=0.1)

Hello, thank you for offering nice repository.
I wonder whether you can additionally offer code of your baseline. (Self-training only)

Hello, would it be convenient for you to provide the pretrained model trained with SYNTHIA-only ?