dvl-tum / ciagan Goto Github PK

Great paper and great work, I'd like to ask you for the evaluation script used to generate the metrics used in the paper, in order to be able to compute them for comparison.

Thank you!

Great paper and great work, I'd like to ask you for the evaluation script used to generate the metrics used in the paper, in order to be able to compute them for comparison.

Thank you!

Hello , can anyone know how to fix this error :
RuntimeError: Error(s) in loading state_dict for Generator:
size mismatch for encode_one_hot.0.weight: copying a param with shape torch.Size([256, 5]) from checkpoint, the shape in current model is torch.Size([256, 1200]).

RuntimeError: Given groups=1, weight of size [3, 16, 3, 3], expected input[1, 32, 64, 64] to have 16 channels, but got 32 channels instead

Hi, I want to save the anonymous images generated by the pre-training model, I guess that the "save_images" function in the "train.py" file does this, but I don't know what the parameters "out, GT, target, INP" represent respectively. Can you provide more details? It will be better if you could give a code example of how a pre-trained model generates images!

请问为什么训练后没有出来模型？

Hi I enjoyed the paper and was able to run the training example. But what is the next step to actually 'run the code' on an image or video? Thanks!

Could you tell me how can I use CIAGAN in video?

Hi, thank you for sharing your code!
Do you have any plan to share your pre-trained models?

So, I downloaded the pre-trained model (link in README).

• Working directory /ciagan/source

1. Created a folder 0 inside mydata folder and added their my photos (000001.jpg, 000002.jpg ...)

2. Created landmarks using this line:
   python process_data.py --input ../mydata --output ../output/

3. And the run the test.py file
python test.py --data ../output/ --out ../output

So here are the results:

I mean, one of the advantages of this paper was creating a new identity and another face detection algorithm will probably detect these faces, however the other advantage was realistic results, but the faces I get are a bit intimidating.

Am I doing everything correctly?
Also, how can I insert these new faces into original images?
Any help, pointer or code is appreciated!

Hi, I enjoyed the paper and was able to run the training example.
I set up the CelebA dataset with 1200 identities provided in legit_indices.npy. However, I am not sure about the hyper-parameter setting:

1. The learning rate is 0.0001 in train.py but 0.00001 in the paper(section 4.1)
2. The iterations of the critic, generator, and siamese are 5, 1, 1 in train.py, and 1, 3, 1 in run_training.py.
3. filter number, batch size, etc

Could you give me the detailed hyper-parameters for training on the dataset with 1200 identities? Thanks!

The current hyper-parameters I'm using, which yield an unsatisfactory result:
'TRAIN_PARAMS': {
'ARCH_NUM': 'unet_flex',
'ARCH_SIAM': 'resnet_siam',
'EPOCH_START': 0,
'EPOCHS_NUM': 120,
'LEARNING_RATE': 0.00001,
'FILTER_NUM': 32,
'ITER_CRITIC': 1,
'ITER_GENERATOR': 3,
'ITER_SIAMESE': 1,
'GAN_TYPE': 'lsgan',
'FLAG_SIAM_MASK': False,
},
'DATA_PARAMS':{
'LABEL_NUM': 1200,
'WORKERS_NUM': 4,
'BATCH_SIZE': 32,
'IMG_SIZE': 128,
'FLAG_DATA_AUGM': True,
},
'OUTPUT_PARAMS': {
'SAVE_EPOCH': 1,
'SAVE_CHECKPOINT': 60,
'LOG_ITER': 2,
'COMMENT': "Something here",
'EXP_TRY': 'check',
}

How do I use the legit_indices file to obtain the train/test split? As I got it they should be 1200 for training and 363 for testing, but the split is not stated neither here nor in the original paper