Stage 3 of Pandora

This commit is contained in:
mokad6 committed 2024-01-18 09:36:32 -05:00
1 parent 783613d0dc
commit 54d5bff351
3 files changed
+62 -10

No files matched your search

@@ -0,0 +1,47 @@
import argparse
import os
from tqdm import tqdm
from pathlib import Path
import json
def load_from_json(filename: Path):
"""Load a dictionary from a JSON filename.
Args:
filename: The filename to load from.
"""
assert filename.suffix == ".json"
with open(filename, encoding="UTF-8") as file:
return json.load(file)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--input_json', type=str)
parser.add_argument('--output_json', type=str)
parser.add_argument('--img_addr', type=str)
args = parser.parse_args()
assert os.path.isfile(args.input_json), 'The input JSON file is missing!'
old_json_data = load_from_json( Path(args.input_json) )
new_json_data = {}
all_keys = old_json_data.keys() - {'frames'}
for key in all_keys:
new_json_data[key] = old_json_data[key]
frames = []
for frame in tqdm(old_json_data['frames']):
base = str(Path(frame["file_path"]).stem)
if Path(args.img_addr + '/' + base + '.exr').is_file():
new_frame = {
"file_path": frame["file_path"],
"transform_matrix": frame["transform_matrix"],
}
frames.append(new_frame)
new_json_data['frames'] = frames
with open(args.output_json, "w") as outfile:
json.dump(new_json_data, outfile)
print("All done!")
+8 -7
View File
@@ -320,13 +320,14 @@ class LanternModel(NerfactoModel):
gt_image=image,
)
# MSE loss for mask with weights
mask_in_float = batch['mask'].type(torch.float).to(self.device)
negative_mask_in_float = torch.ones(mask_in_float.shape).to(self.device) - mask_in_float
loss_mask_fast_expo = (mask1_w * (( negative_mask_in_float - torch.unsqueeze(outputs["validity_f"], 1)) ** 2)).mean()
loss_mask_well_expo = (mask2_w * (( mask_in_float - torch.unsqueeze(outputs["validity_w"], 1)) ** 2)).mean()
loss_dict["validity_loss_well_exposed"] = loss_mask_well_expo
loss_dict["validity_loss_fast_exposure"] = loss_mask_fast_expo
if "mask" in batch:
# MSE loss for mask with weights
mask_in_float = batch['mask'].type(torch.float).to(self.device)
negative_mask_in_float = torch.ones(mask_in_float.shape).to(self.device) - mask_in_float
loss_mask_fast_expo = (mask1_w * (( negative_mask_in_float - torch.unsqueeze(outputs["validity_f"], 1)) ** 2)).mean()
loss_mask_well_expo = (mask2_w * (( mask_in_float - torch.unsqueeze(outputs["validity_w"], 1)) ** 2)).mean()
loss_dict["validity_loss_well_exposed"] = loss_mask_well_expo
loss_dict["validity_loss_fast_exposure"] = loss_mask_fast_expo
if weights_for_RGB_loss is not None:
loss_dict["rgb_loss"] = (weights_for_RGB_loss * ((gt_rgb - pred_rgb) ** 2)).mean()
+7 -3
View File
@@ -549,16 +549,20 @@ class RenderInterpolated(BaseRender):
install_checks.check_ffmpeg_installed()
output_names = []
if self.pose_source == "eval":
assert pipeline.datamanager.eval_dataset is not None
cameras = pipeline.datamanager.eval_dataset.cameras
for file_name in pipeline.datamanager.eval_dataset.image_filenames:
output_names.append(file_name.stem)
else:
assert pipeline.datamanager.train_dataset is not None
cameras = pipeline.datamanager.train_dataset.cameras
for file_name in pipeline.datamanager.train_dataset.image_filenames:
output_names.append(file_name.stem)
output_names = []
for file_name in pipeline.datamanager.train_dataset.image_filenames:
output_names.append(file_name.stem)
seconds = self.interpolation_steps * len(cameras) / self.frame_rate
camera_path = get_interpolated_camera_path(
cameras=cameras,