No jobs
Back to SD 2.1. Mari

Train: Metric Depth V8 a4c5aab1

Supersedes Train: Metric Depth V7 by wiring the unified euler-loading intrinsics ingestion pathway. data.sources.{rds,vkitti2}.dataset_kwargs.intrinsics_root is loaded as an euler-loading intrinsics modality and bridged into the existing camera transform chain (AttachCameraMetadata -> crop/resize/flip -> FinalizeCameraSample), replacing ad-hoc sidecar intrinsics. Enables real intrinsics by default: data.camera.enabled=true, metadata_format=euler_loading, require_for_training=true, intrinsics_loading=hierarchical (one shared calibration file per scene/camera), intrinsics_metadata_scope=intrinsics. Existing sidecar/constant/approximate camera modes remain available; geometry losses and the model geometry branch stay disabled (model.use_geometry=false).

training cancelled

Output

pending a4c5aab1
Launch Output
Waiting for SLURM job to start...

Datasets

5
Dataset clear_root_rds
Type rgb
Split full
Path /cluster/work/igp_psr/drothenpiele/data/rds/rgb.zip/
Dataset depth_root_rds
Type depth
Split full
Path /cluster/work/igp_psr/drothenpiele/data/rds/depth.zip/
Dataset intrinsics_root_rds
Type intrinsics
Split full
Path /cluster/work/igp_psr/drothenpiele/data/rds/calibration.zip/
Dataset train_hazy_root_rds
Type rgb
Split train
Path /cluster/work/igp_psr/drothenpiele/data/rds/gloomy/noise_model/foggy_rgb.zip/
Dataset val_hazy_root_rds
Type rgb
Split val
Path /cluster/work/igp_psr/drothenpiele/data/rds/gloomy/noise_model/foggy_rgb.zip/

Execution Artifacts

2
run.sh

Published Exports

Launch-Owned Exports

These semantic handles are what downstream pipelines should resolve against. Auto-published exports come from config-template metadata captured on this launch.

Published

This launch does not publish any exports yet.

Raw Artifacts

Parameters

257

Typed Parameters

Dataset clear_root_rds
real-drive-sim / full (4)
Dataset depth_root_rds
real-drive-sim / full (5)
Dataset intrinsics_root_rds
real-drive-sim / full (6)
Dataset train_hazy_root_rds
real-drive-sim / train (103)
Dataset val_hazy_root_rds
real-drive-sim / val (104)
Dataset clear_root_vkitti2
real-drive-sim / full (4)
Dataset depth_root_vkitti2
real-drive-sim / full (4)
Dataset intrinsics_root_vkitti2
real-drive-sim / full (4)
Dataset train_hazy_vkitti
real-drive-sim / full (4)
Dataset val_hazy_vkitti
real-drive-sim / full (4)

Simple Parameters

clip
true
crop
center
seed
42
tracker
wandb
use_ema
false
gpu_type
rtx_4090:1
job_name
sd_train_concat
norm_max
1
norm_min
-1
run_name
joint-dehaze-metric
tmp_size
10G
adam_8bit
true
data_kind
real_drive_sim
ema_decay
0.9999
ema_dtype
float32
log_every
10
max_depth
800
min_depth
1
adam_beta1
0.9
adam_beta2
0.999
batch_size
16
clear_root
4
depth_mode
metric_log
depth_root
5
image_size
[384, 768]
log_images
true
num_epochs
200
output_dir
/cluster/scratch/drothenpiele/SD21/exp_1
time_limit
2-00:00:00
mem_per_cpu
8G
num_workers
4
adam_epsilon
1e-8
adam_foreach
false
aspect_ratio
[1, 2]
camera_model
pinhole
conditioning
concat
lambda_depth
1
lambda_point
0.0
min_lr_ratio
0.01
project_name
joint-dehazing
use_geometry
true
warmup_ratio
0.05
weight_decay
0
cpus_per_task
8
lambda_camera
0.1
lambda_dehaze
1
lambda_eg_ssi
0.0
learning_rate
0.00004
max_depth_rds
800
max_grad_norm
1
min_depth_rds
0.00001
val_hazy_root
104
camera_enabled
true
freeze_encoder
false
log_visibility
true
num_log_images
4
use_depth_head
true
val_batch_size
10
camera_fallback
invalid
depth_head_type
base_residual
enable_xformers
true
euler_train_dir
/cluster/work/igp_psr/drothenpiele/data/out/train/sd21-joint
lambda_depth_tv
0.0002
mixed_precision
fp16
prediction_type
v_prediction
source_kind_rds
real_drive_sim
train_hazy_root
103
visibility_cmap
viridis
camera_embedding
sine
camera_emit_rays
false
camera_mlp_ratio
4
camera_num_heads
4
camera_token_dim
256
depth_noise_type
annealed_multires
lambda_log_depth
0.0
lr_schedule_type
constant_with_warmup
min_max_quantile
0.02
no_decay_enabled
false
pretrained_model
sd2-community/stable-diffusion-2-1
sky_depth_meters
800
camera_num_layers
2
camera_num_tokens
4
conv_in_init_mode
marigold
geometry_lambda_z
0.15
geometry_timestep
0
joint_weight_psnr
1
joint_weight_ssim
10
lambda_depth_head
0.2
lambda_lambda_mse
0.0
lambda_visibility
0.05
max_depth_vkitti2
300
min_depth_vkitti2
0.00001
sampling_strategy
source_balanced
source_weight_rds
1
depth_ensemble_tol
0.001
lambda_uncertainty
0.0
model_depth_output
depth_head
val_every_n_epochs
2
camera_feature_pool
avg
depth_ensemble_size
1
geometry_lambda_phi
1
num_inference_steps
25
sampling_epoch_size
balanced
save_every_n_epochs
2
source_kind_vkitti2
vkitti2
use_visibility_head
true
warmup_start_factor
0.001
camera_metadata_root
null
external_camera_mode
none
geometry_loss_robust
charbonnier
lambda_vae_depth_rec
0.0
lambda_visibility_tv
0.005
sky_depth_meters_rds
800
camera_branch_enabled
true
camera_embedding_size
latent
depth_gradient_robust
charbonnier
depth_multires_levels
4
encoder_learning_rate
0.000015
geometry_head_enabled
true
geometry_lambda_theta
1
lambda_depth_gradient
0.05
lambda_geometry_total
1.0
source_weight_vkitti2
1
use_cross_task_fusion
false
vae_decoder_trainable
false
visibility_rank_pairs
128
weight_decay_backbone
0
weight_decay_no_decay
0
zero_grad_set_to_none
true
camera_feature_sources
auto
camera_metadata_format
euler_loading
depth_ensemble_max_res
1024
depth_smoothness_space
depth_head
freeze_cross_attention
true
gradient_checkpointing
true
inference_depth_output
depth_head
lambda_geo_consistency
0.0
lambda_visibility_rank
0
visibility_rank_margin
0.05
camera_embedding_stride
8
camera_prediction_space
residual_intrinsics
depth_ensemble_max_iter
2
depth_multires_strength
0.9
depth_smoothness_scales
4
geometric_pairs_enabled
false
joint_weight_delta1_pct
0.5
keep_last_n_checkpoints
3
validation_depth_output
both
visibility_target_gamma
4
visibility_warmup_steps
1000
cross_task_fusion_blocks
[0]
cross_task_fusion_detach
true
depth_ensemble_reduction
median
depth_normalization_type
log_depth
geometry_camera_encoding
sine
geometry_head_num_blocks
2
joint_weight_abs_rel_pct
0.5
sky_depth_meters_vkitti2
800
camera_intrinsics_loading
hierarchical
depth_diagnostics_enabled
true
depth_head_residual_scale
0.05
geometry_depth_convention
z_depth
geometry_metric_depth_max
800
geometry_metric_depth_min
1
num_inference_steps_depth
20
vae_decoder_learning_rate
0.000001
visibility_apply_to_depth
false
camera_residual_logit_clip
4
depth_head_hidden_channels
64
depth_resize_interpolation
bilinear
depth_smoothness_normalize
true
detach_rays_for_point_loss
true
enable_efficient_attention
true
lambda_visibility_preserve
0.05
visibility_apply_to_dehaze
true
visibility_apply_to_fusion
false
visibility_hidden_channels
32
visibility_target_quantile
0.9
weight_decay_depth_decoder
0
weight_decay_geometry_head
0
camera_require_for_training
true
camera_sine_num_frequencies
null
checkpoint_selection_metric
"joint" # "joint" or single metric: psnr/ssim/delta1/abs_rel/val/...
depth_decoder_learning_rate
0.00004
emit_camera_full_resolution
true
geometry_eg_ssi_edge_source
hazy
geometry_encoder_input_mode
hazy_hazy
geometry_head_learning_rate
0.00004
geometry_output_uncertainty
true
geometry_ray_representation
unit
gradient_accumulation_steps
1
validation_geometry_enabled
true
weight_decay_dehaze_decoder
0
camera_output_representation
angles
camera_valid_for_approximate
false
cross_task_fusion_directions
["rgb_to_depth"]
dehaze_decoder_learning_rate
0.00004
depth_head_context_dilations
[1, 2, 4]
depth_smoothness_edge_source
clear
depth_smoothness_edge_weight
10
geometry_head_residual_scale
0.05
lambda_depth_edge_smoothness
0.002
weight_decay_geometry_camera
0
depth_diagnostics_edge_source
dehazed
depth_diagnostics_edge_weight
10
depth_head_base_upsample_mode
bilinear
geometry_camera_learning_rate
0.00004
geometry_head_hidden_channels
128
geometry_sine_num_frequencies
64
visibility_decoder_grad_scale
1.0
camera_approximate_fov_degrees
60
checkpoint_selection_direction
"auto" # used for single-metric selection: auto|max|min
cross_task_fusion_learned_gate
true
geometry_use_deterministic_vae
true
depth_multires_downscale_factor
2
lambda_depth_gt_edge_smoothness
0
validation_geometry_camera_mode
predicted
validation_geometry_output_mode
geometry_head
camera_include_input_in_encoding
null
camera_intrinsics_metadata_scope
intrinsics
cross_task_fusion_gate_init_bias
-4
cross_task_fusion_sender_lowpass
false
depth_head_use_depthwise_context
true
geometric_pairs_views_per_sample
2
geometry_condition_on_visibility
false
geometry_detach_encoder_features
true
cross_task_fusion_hidden_channels
64
depth_diagnostics_normalize_depth
true
geometric_pairs_emit_pair_metadata
true
geometry_include_input_in_encoding
false
lambda_depth_multiscale_smoothness
0.001
depth_ensemble_regularizer_strength
0.02
depth_diagnostics_road_edge_threshold
0.08
geometry_head_detach_camera_embedding
true
depth_diagnostics_road_bottom_fraction
0.4
geometry_encoder_no_grad_when_detached
true
cross_task_fusion_sender_lowpass_kernel
3
depth_diagnostics_high_frequency_kernel
5
validation_geometry_compute_edge_metrics
true
validation_geometry_compute_point_metrics
false
validation_geometry_max_points_per_sample
20000
validation_geometry_save_geometry_samples
true
validation_geometry_compute_camera_metrics
true
validation_geometry_compute_resolution_sweep
false
validation_geometry_max_point_metric_samples
8
validation_geometry_save_point_cloud_samples
false
validation_geometry_compute_uncertainty_metrics
true
Raw JSON
{
  "clip": "true",
  "crop": "center",
  "seed": 42,
  "tracker": "wandb",
  "use_ema": "false",
  "gpu_type": "rtx_4090:1",
  "job_name": "sd_train_concat",
  "norm_max": 1,
  "norm_min": -1,
  "run_name": "joint-dehaze-metric",
  "tmp_si...

Events

Launch Events

0
No launch events recorded.
Euler View - ML Experiment Monitor