Edge2car training config.

[zqh0253]

Hi, really nice work! Could you please share the training script for Edge2car?

[zqh0253]

I use the default config to train the shapenet car dataset. However, the training process collapses. This is the generated front image:

Here is my config:

python train.py --outdir=experiment \
--cfg=shapenet --data=/data/cars_128.zip \
--mask_data=/data/shapenet_car_contour.zip \
--data_type=edge \
--semantic_channels=1 \
--render_mask=True \
--dis_mask=True \
--neural_rendering_resolution_initial=128 \
--gpus=3 --batch=12 --mbstd-group=2 \
--gamma=5 --gen_pose_cond=True \
--random_c_prob=0.5 \
--lambda_d_semantic=0.10 \
--lambda_lpips=1 \
--lambda_cross_view=1e-4 \
--only_raw_recons=True \
--wandb_log=True

Is there anything that I have made a mistake with?

[dunbar12138]

Hi, thanks for your interest!

The edge2car training script can be found here.

I also fixed the training code part that might cause the bug you encountered.