Question about SCARED Dataset Structure

[sandokim]

Hello,

Following the instructions provided to obtain the SCARED dataset, I have structured the data as follows:

• RGB images are in /data/scared2019/dataset/dataset_1_keyframe_1/data/rgb_data (e.g., frame_data000000.png, frame_data000001.png, ..., frame_data000196.png).
• Camera poses are in /data/scared2019/dataset/dataset_1_keyframe_1/data/frame_data (e.g., frame_data000000.json, frame_data000001.json, ..., frame_data000196.json).

I am currently working with the SCARED dataset and have a couple of questions:

1. In the scene/pose_optimizer.py file, I see that the PoseModel class loads RGB images from an input folder. Should I rename the rgb_data folder to input to match the expected structure, or is there another approach?

Free-SurGS/scene/pose_optimizer.py

Lines 354 to 363 in 4cd688c

	class PoseModel:
	def __init__(self, args, device="cuda", sample_rate=8):
	self.data_path = args.source_path
	self.data_type = args.data_type
	self.start_frame = args.frame_start
	self.end_frame = args.frame_end
	rgb_paths = glob.glob(os.path.join(self.data_path, 'input', '*.png')) + \
	glob.glob(os.path.join(self.data_path, 'input', '*.jpeg'))+ \
	glob.glob(os.path.join(self.data_path, 'input', '*.jpg'))
	self.W, self.H = Image.open(rgb_paths[0]).size

2. I believe the frame_data folder contains ground truth camera poses in JSON format. Is my understanding correct that the rgb_data images are used for training, and the camera pose errors are calculated using the corresponding JSON files?

Could you clarify the correct data structure and how the inputs should be organized for training?

Thank you!