kailigo / cvczsl Goto Github PK

Hi,

I tried to reproduce the results on AwA1, without any modification on your code and data. However, I got
best_ep: 829, zsl: 0.6889, gzsl: seen=0.7759, unseen=0.5984, h=0.6757
zsl: 0.6889, gzsl: seen=0.7759, unseen=0.5984, h=0.6757
on AwA1 inductive training. Do you have any suggestion to reproduce the results in your paper?

Sorry to disturb you.
I have read your paper, an interesting work, and tried to implement your code, but I have some problems.

1. What is weight_gen_model_name? Does it contain the parameters of the pre-training model? If the answer is yes, is it pre-trained on ImageNet. If not, can you provide some details? If I train the model on other datasets, e.g. CUB, it seems I don't have './models/CUB1/weight_gen_model.pt', where can I get the file?
2. Are the parameters trained on different data sets consistent?
   python trans_train.py --dataset AWA1 --ways 16 --shot 1 --lr 1e-4 --opt_decay 1e-5 --step_size 200 --loss_q 5e-1 --trans_model_name trans_s1w16_lr4_opt5_ss200_q5e1 --log_file trans_s1w16_lr4_opt5_ss200_q5e1

Looking forward to your reply.

(1) Why did you used the feature set from DEM_CVPR2017 instead of dataset from Xian, reported in your paper?
(2) Is there difference when use different feature set?

I am very interested in your paper, but something I am not clear.
Could you tell me when you can upload the code?