Thank you for your great work and open source code. However, I am confused about the code:

1. local variable 'net' referenced before assignment in

   CR-NeRF-PyTorch/models/nerf_decoder_stylenerf.py

   Line 438 in 334fa6b

   hid0=self.feat_upsample_list[idx](net)

2. the parameter "decoder_num_res_blocks" is not being used in:

   CR-NeRF-PyTorch/models/linearStyleTransfer.py

   Line 279 in 334fa6b

   def __init__(self, args, residual_blocks=2):

3. in the article, it is mentioned to set "m" to be smaller than "hcr1hcr2". However, in the code, it appears that the decoder receives image features that are of the same size as the image:

   CR-NeRF-PyTorch/models/linearStyleTransfer.py

   Line 283 in 334fa6b

   self.decoder=NeuralRenderer(img_size=(args.img_wh[0],args.img_wh[1]) , featmap_size=(args.img_wh[0],args.img_wh[1]), feat_nc=args.nerf_out_dim, out_dim=3, args_here=args )

4. moreover, may I ask if I am correct in understanding that when using a decoder, the loss can only be calculated for the entire image, since the decoder generates the complete image every time, and cannot sample rays like traditional NERF to calculate the loss on discrete rays?

Hi, thanks for your interesting work.

Could you provide more details about the process of eq (9)? Specifically, how to apply the grid sampling strategy (GS) here?

Thanks a lot.

Hi, I tried to retrain the model according to the following command, and the test psnr was only 25.33. In addition, I used the CR-NeRF-branden.ckpt model in the cloud disk for testing, and the accuracy was only 24.55. Can you please help me figure out where the problem is with the settings?

Training
python train_mask_grid_sample.py \
	--root_dir ../brandenburg_gate/ \
	--dataset_name phototourism \
	--save_dir ./tmp \
	--img_downscale 2 \
	--N_importance 64 --N_samples 64 \
	--num_epochs 1 --batch_size 1024 \
	--optimizer adam --lr 5e-4 --lr_scheduler cosine \
	--N_emb_xyz 15 --N_vocab 1500 \
	--maskrs_max 5e-2 --maskrs_min 6e-3 \
	--maskrs_k 1e-3 --maskrd 0 --N_a 48 \
	--weightKL 1e-5 --weightRecA 1e-3 \
	--weightMS 1e-6 --chunk 1310720 \
	--encode_a --encode_c  --encode_random \
	--model_mode 1-1 --decoder linearStyle \
	--decoder_num_res_blocks 1 \
	--nerf_out_dim 64 \
	--use_cache --proj_name style_gnerf \
	--use_mask --exp_name train/exp1 \
	--num_gpus 4

Testing
python eval.py \
  --root_dir ../brandenburg_gate/ \
  --save_dir ./tmp \
  --dataset_name phototourism \
  --scene_name train/exp1 \
  --split test_test \
  --img_downscale 2 \
  --N_samples 256 \
  --N_importance 256 \
  --N_emb_xyz 15 \
  --N_vocab 1500 \
  --ckpt_path ./tmp/ckpts/train/exp1/epoch20.ckpt \
  --chunk 20480 --img_wh 320 240 \
  --encode_a --decoder linearStyle \
  --decoder_num_res_blocks 1 \
  --nerf_out_dim 64

Can u tell me how to train on blender dataset? i really dont get it. Big thanks for help!!

Realization of multiple 2D images to build a 2D image, fusing the elements of multiple images into a single image, can the project be done?
实现多张二维图片构建一张二维图片，将多张图片的要素融合到一张图片上，该项目能办到吗？

Hi YifYang,
Thanks for your great work, I've just read your paper and don't read your code yet, I don't understand some questions:

1. How to choose m rays during training, choose m pixels in a single image randomly, or m neighbors of a pixel? or some else rule? A set of m rays' features match m pixels' colors in loss function or a single pixel's color?
2. In eq(5),why there need the second term (the beta term). And why it can be modeld as the D_theta5 network.
   Thanks a lot.