def save_result(self, i, batch_data, out_depth):
b = batch_data['left_im'].size(0)
for ii in range(b):
curr_im_name = self.output_dir + ('%05d' %
i) + '_' + str(ii) + '.png'
depth = out_depth[ii, :, :]
depth = 1.0 / depth
depth_3 = self.gray2jet(depth)
torchvision.utils.save_image(depth_3, curr_im_name)
def gray2jet(self, dmap):
cmap = plt.get_cmap('magma')
if len(dmap.size()) == 4:
dmap_0 = dmap[0, 0, :, :].cpu().numpy()
elif len(dmap.size()) == 3:
dmap_0 = dmap[0, :].cpu().numpy()
elif len(dmap.size()) == 2:
dmap_0 = dmap.cpu().numpy()
else:
raise 'Wrong dimensions of depth: {}'.format(dmap.size())
dmap_norm = (dmap_0 - dmap_0.min()) / (dmap_0.max() - dmap_0.min())
dmap_col = cmap(dmap_norm)
dmap_col = dmap_col[:, :, 0:3]
dmap_col = np.transpose(dmap_col, (2, 0, 1))
return torch.tensor(dmap_col).float().to(self.device)
if __name__ == '__main__':
opts = Options().opts
TestFaster(opts)
def get_online_train_categories(self):
model_names = sorted(x for x in os.listdir(self.eval_dir) if x.endswith('.pth'))
model_names = [x for x in model_names if 'Disp' in x]
model_tags = [x.split('Disp')[0] for x in model_names]
if self.dataset_tag == 'nyu':
model_tags = [x[3:-1] for x in model_tags]
model_tags = [x[:-1] if x.endswith('_') else x for x in model_tags]
elif self.dataset_tag == 'kitti':
model_tags = [x.split('kitti_')[1][:-1] if 'kitti_' in x else x for x in model_tags]
model_tags = [x.split('_')[0] for x in model_tags]
return model_tags
if __name__ == '__main__':
opts = Options().opts
if opts.dataset_tag == 'nyu':
train_opts = OptionsPretrain().opts
train_opts.root = '/hdd/local/sdb/umar/nyu_indoor/rectified_nyu/'
# getting the pretrain
nyu_cat_file_name = train_opts.nyu_cat_file_name
train_index_end = train_opts.nyu_train_index_end
nyu_cat_name_dict = np.load(nyu_cat_file_name, allow_pickle=True).item()
categories = sorted(nyu_cat_name_dict.keys())
train_categories = categories[:categories.index(train_index_end)]
opts.nyu_pretrain_categories = [x[:-1] for x in train_categories]
# print(opts.nyu_pretrain_categories)
Evaluator = EvaluateResults(opts)
print('NYU results: {}'.format(Evaluator.complete_evaluation()))
elif opts.dataset_tag == 'kitti':
from dir_options.test_options import Options
from test_directory import EvalDirectory
import time
import os
if __name__ == '__main__':
opts = Options().opts
runs = opts.runs
list_eval_dir_kitti = ['trained_models/online_models_kitti/']
list_eval_dir_nyu = ['trained_models/online_models_nyu/']
list_results_dir_kitti = ['results/online_test_loss/kitti_online/']
list_results_dir_nyu = ['results/online_test_loss/nyu_online/']
for run in runs:
list_eval_dir_kitti.append('trained_models/online_models_kitti_replay_reg_run' + run + '/')
list_eval_dir_nyu.append('trained_models/online_models_nyu_replay_reg_run' + run + '/')
list_results_dir_kitti.append('results/replay_reg_test_loss_run' + run + '/kitti_online/')
list_results_dir_nyu.append('results/replay_reg_test_loss_run' + run + '/nyu_online/')
assert len(list_eval_dir_kitti) == \
len(list_eval_dir_nyu) == \
len(list_results_dir_kitti) == \
len(list_results_dir_nyu), 'Check the number of elements'
for i in range(len(list_eval_dir_kitti)):
metric_mean = {}
metric_std = {}
for metric in metrics:
metric_mean[metric] = np.mean(metric_lists[metric])
metric_std[metric] = np.std(metric_lists[metric])
print(metric_std[metric])
data = np.reshape(np.array(list(metric_mean.values())), (1, -1))
df = pd.DataFrame(data)
print(df.to_latex(float_format="%0.4f"))
data = np.reshape(np.array(metric_std.values()), (1, -1))
df = pd.DataFrame(data)
print(df.to_latex(float_format="%0.6f"))
if __name__ == '__main__':
opts = Options().opts
metrics = opts.metrics
models = [
'trained_models/online_models_vkitti/05_Scene20_Disp_000_15287.pth',
'trained_models/online_models_vkitti_replay_reg_run1/05_Scene20_Disp_000_15287.pth',
'trained_models/online_models_vkitti_replay_reg_run2/05_Scene20_Disp_000_15287.pth',
'trained_models/online_models_vkitti_replay_reg_run3/05_Scene20_Disp_000_15287.pth',
'trained_models/online_models_vkitti_fog_rain/05_Scene20_Disp_000_19109.pth',
'trained_models/online_models_vkitti_fog_rain_replay_reg_run1/05_Scene20_Disp_000_19109.pth',
'trained_models/online_models_vkitti_fog_rain_replay_reg_run2/05_Scene20_Disp_000_19109.pth',
'trained_models/online_models_vkitti_fog_rain_replay_reg_run3/05_Scene20_Disp_000_19109.pth'
]
tags = ['kitti']
kitti_dataset = Datasets.KittiDepthTestDataset(opts)
kitti_dataloader = data.DataLoader(kitti_dataset,
def display_latex_style_w_std(mean, std):
h = len(mean)
w = len(mean[0])
for r in range(h):
for c in range(w):
mean_str = f"{mean[r][c]:0.4f}"
std_str = f"{std[r][c]:0.4f}"
complete_string = mean_str + ' + ' + std_str + ' & '
print(complete_string, end="")
print('\\\\')
if __name__ == '__main__':
opts = Options().opts
pre_opts = OptionsPretrain().opts
metrics = opts.metrics
runs = opts.runs
nyu_train_index_end = pre_opts.nyu_train_index_end
nyu_categories = list(NYUCategorySplit(pre_opts).__call__().keys())
nyu_pretrain_categories = nyu_categories[:nyu_categories.
index(nyu_train_index_end)]
opts.nyu_pretrain_categories = [x[:-1] for x in nyu_pretrain_categories]
list_eval_dir_kitti = ['trained_models/online_models_kitti/']
list_eval_dir_nyu = ['trained_models/online_models_nyu/']
list_results_dir_kitti = ['results/online_test_loss/kitti_online/']
list_results_dir_nyu = ['results/online_test_loss/nyu_online/']
for run in runs:
import torch.utils.data as data
import os
from test import TestFaster
from dir_options.test_options import Options
from dir_dataset import Datasets
if __name__ == '__main__':
opts = Options().opts
opts.qual_results = True
kitti_online_model_dir = 'trained_models/online_models_kitti/'
kitti_rep_reg_model_dir = 'trained_models/online_models_kitti_rep_reg_run3/'
vkitti_online_model_dir = 'trained_models/online_models_vkitti/'
vkitti_rep_reg_model_dir = 'trained_models/online_models_vkitti_rep_reg_run3/'
kitti_online_models = sorted(
[x for x in os.listdir(kitti_online_model_dir) if x.endswith('.pth')])
kitti_rep_reg_models = sorted(
[x for x in os.listdir(kitti_rep_reg_model_dir) if x.endswith('.pth')])
vkitti_online_models = sorted(
[x for x in os.listdir(vkitti_online_model_dir) if x.endswith('.pth')])
vkitti_rep_reg_models = sorted(
[x for x in os.listdir(vkitti_online_model_dir) if x.endswith('.pth')])
kitti_online_model = kitti_online_model_dir + kitti_online_models[-1]
kitti_rep_reg_model = kitti_rep_reg_model_dir + kitti_rep_reg_models[-1]
vkitti_online_model = vkitti_online_model_dir + vkitti_online_models[-1]
vkitti_rep_reg_model = vkitti_rep_reg_model_dir + vkitti_rep_reg_models[-1]
model_paths = [
kitti_online_model, kitti_rep_reg_model, vkitti_online_model,
vkitti_rep_reg_model
]
def display_latex_style_w_std(mean, std):
h = len(mean)
w = len(mean[0])
for r in range(h):
for c in range(w):
mean_str = f"{mean[r][c]:0.4f}"
std_str = f"{std[r][c]:0.4f}"
complete_string = mean_str + ' + ' + std_str + ' & '
print(complete_string, end="")
print('\\\\')
if __name__ == '__main__':
opts = Options().opts
metrics = opts.metrics
runs = opts.runs
list_eval_dir_kitti = ['trained_models/online_models_kitti/']
list_eval_dir_vkitti = ['trained_models/online_models_vkitti/']
list_results_dir_kitti = ['results/online_test_loss/kitti_online/']
list_results_dir_vkitti = ['results/online_test_loss/vkitti_online/']
for run in runs:
list_eval_dir_kitti.append(
'trained_models/online_models_kitti_rep_reg_run' + run + '/')
list_eval_dir_vkitti.append(
'trained_models/online_models_vkitti_rep_reg_run' + run + '/')
list_results_dir_kitti.append('results/rep_reg_online_test_loss_run' +
run + '/kitti_online/')