[street] The image synthesized during training is black

[songyn95]

https://github.com/NVlabs/few-shot-vid2vid/

I am going to synthesize traffic video(2 sequences in the training set, 160 images for each). During the training, I found that the synthetic image was black during iteration 14/70. What is the reason and how to solve it? The training and test commands are as follows:

1. python train.py --name test --dataset_mode fewshot_street --adaptive_spade --loadSize 512 --fineSize 512 --batchSize 2 --dataroot datasets --label_nc 0 --gpu_ids 2
2. python test.py --name test --dataset_mode fewshot_street --adaptive_spade --loadSize 512 --fineSize 512 --seq_path datasets/A_01 --ref_img_path datasets/B_01 --label_nc 0 --gpu_ids 3

[YukunZEX]

i got same problem in old version of fsvid2vid project, have you solve it?