GeoBrain/blending-noise-suppression

Blending Noise Suppression Benchmark

This task performs supervised deblending on paired common-receiver gathers.

Task

The model directly learns a mapping from pseudo-deblended input to the cleaner forwarding reference:

denoised = model(pseudo_deblended_input)

This is a paired regression task, not a synthetic noise injection task.

Dataset

Current paired SEG-Y inputs:

• input: /root/Desktop/data/T02_pseudo_deblended_common_receiver_mod.sgy
• target: /root/Desktop/data/T02_forwarding_acoustic_common_receiver.sgy

Logical volume shape:

• (386, 270, 2000)
  • 386 common-receiver gathers
  • 270 traces per common-receiver gather
  • 2000 time samples per trace

Split policy:

• sequential receiver-gather split
• train: 300
• val: 43
• test: 43

Models

• unet
• res_unet
• dncnn
• atten_unet

All model definitions are loaded from model/blending_noise_suppression/.

Preprocessing

• paired loading with shape consistency check
• shared normalization statistics for input and target
• overlapping 2D patches of size 128 x 256
• overlap ratio 0.5

Scripts

Training:

• bash scripts/blending_noise_suppression/train_denoise_unet.sh
• bash scripts/blending_noise_suppression/train_denoise_res_unet.sh
• bash scripts/blending_noise_suppression/train_denoise_dncnn.sh
• bash scripts/blending_noise_suppression/train_denoise_atten_unet.sh

Each training script now runs a fixed three-seed sweep:

• seeds: 42, 43, 44
• run names: <experiment.name>_<dataset_id>_<noise_tag>_seed<seed> For the current input file T02_pseudo_deblended_common_receiver_mod.sgy, this becomes ..._T02_mod_seed<seed>.
• checkpoints: /root/Desktop/data/results/blending_noise_suppression/<model>/<run_name>/checkpoints/

Inference:

• bash scripts/blending_noise_suppression/inference_denoise_unet.sh
• bash scripts/blending_noise_suppression/inference_denoise_res_unet.sh
• bash scripts/blending_noise_suppression/inference_denoise_dncnn.sh
• bash scripts/blending_noise_suppression/inference_denoise_atten_unet.sh

Each inference script matches the same three seeds, writes per-seed outputs to:

• /root/Desktop/data/results/blending_noise_suppression/<model>/<run_name>/inference

and also saves a cross-seed aggregate summary to:

• /root/Desktop/data/results/blending_noise_suppression/<model>/<experiment.name>_<dataset_id>_<noise_tag>_seed_stats/metrics_summary_mean_std.json

Run all:

• bash scripts/blending_noise_suppression/run_all_blending_models.sh

Upload

Dataset upload helpers in this folder are:

• bash scripts/blending_noise_suppression/upload_dataset_to_hf.sh

This uploads:

• input/T02_pseudo_deblended_common_receiver_mod.sgy
• target/T02_forwarding_acoustic_common_receiver.sgy
• assets/deblending/*.png from scripts/blending_noise_suppression/assets/
• generated dataset card README.md

The current shell wrappers follow the same style as the random-noise uploader:

• edit the fixed options directly inside the .sh file
• then run the shell script

Model upload helper in this folder:

• bash scripts/blending_noise_suppression/upload_blending_model_to_hf.sh

This uploads:

• models/<model>/<dataset_variant>_seed<seed>/best.pt
• models/<model>/<dataset_variant>_seed<seed>/config.yaml
• generated model card README.md

Outputs

Each inference directory contains:

• metrics_per_gather.csv
• metrics_per_sample.csv
• metrics_summary.json
• visualizations/
• optional npy/

Inference defaults aligned with the random-noise workflow:

• n_viz_gathers: 5
• save_npy: false
• per-model inference batch_size matches the corresponding training config

Metrics are reported for:

• input vs target
• denoised vs target
• delta = denoised minus input

For naming consistency with the random-noise workflow:

• metrics_summary.json uses input, denoised, and delta
• the file also keeps a backward-compatible restored alias
• metrics_per_sample.csv provides a generic sample-level view, while metrics_per_gather.csv keeps the correct gather-specific naming

Results

Mean +- std over available seeds, computed from *_seed_stats/metrics_summary_mean_std.json. Metrics are reported on common-receiver gathers in the normalized domain.

Dataset Variant T02_comp

Method	Parameters (M)	SNR	PSNR	SSIM	MAE	MSE	RMSE	EB_WSE_MEDIUM_40_70_NE	EB_WSE_MEDIUM_40_70_SNR	EB_WSE_STRONG_70_100_NE	EB_WSE_STRONG_70_100_SNR	EB_WSE_VERY_WEAK_5_20_NE	EB_WSE_VERY_WEAK_5_20_SNR	EB_WSE_WEAK_20_40_NE	EB_WSE_WEAK_20_40_SNR	FB_FRE_HIGH_ENERGY_RATIO	FB_FRE_HIGH_FREQUENCY_RANGE_HZ	FB_FRE_HIGH_NE	FB_FRE_HIGH_SNR	FB_FRE_LOW_ENERGY_RATIO	FB_FRE_LOW_FREQUENCY_RANGE_HZ	FB_FRE_LOW_NE	FB_FRE_LOW_SNR	FB_FRE_MID_ENERGY_RATIO	FB_FRE_MID_FREQUENCY_RANGE_HZ	FB_FRE_MID_NE	FB_FRE_MID_SNR	FB_FRE_VERY_HIGH_ENERGY_RATIO	FB_FRE_VERY_HIGH_FREQUENCY_RANGE_HZ	FB_FRE_VERY_HIGH_NE	FB_FRE_VERY_HIGH_SNR
Input	-	1.1666+-0.0000	24.1733+-0.0000	0.6475+-0.0000	0.020175+-0.000000	0.003923+-0.000000	0.062252+-0.000000	2.424931+-0.000000	-7.5937+-0.0000	0.354534+-0.000000	9.0088+-0.0000	41.080699+-0.000000	-32.1112+-0.0000	13.037582+-0.000000	-22.1041+-0.0000	0.243824+-0.000000	28.75-41.8	0.873002+-0.000000	1.1802+-0.0000	0.023417+-0.000000	7-15.7	0.898679+-0.000000	0.9282+-0.0000	0.619438+-0.000000	15.7-28.75	0.871390+-0.000000	1.1961+-0.0000	0.006539+-0.000000	41.8-50.5	0.900242+-0.000000	0.9140+-0.0000
UNet	7.76	13.7153+-0.0285	36.7220+-0.0286	0.9659+-0.0011	0.005862+-0.000223	0.000217+-0.000001	0.014654+-0.000046	0.548360+-0.013994	5.2264+-0.2239	0.187143+-0.000813	14.5646+-0.0368	2.023152+-0.118065	-6.0304+-0.5009	1.022692+-0.006941	-0.1832+-0.0606	0.243824+-0.000000	28.75-41.8	0.191022+-0.000523	14.3887+-0.0221	0.023417+-0.000000	7-15.7	0.310998+-0.004445	10.1490+-0.1246	0.619438+-0.000000	15.7-28.75	0.195605+-0.001797	14.1894+-0.0790	0.006539+-0.000000	41.8-50.5	0.354464+-0.016425	9.0312+-0.3973
DnCNN	0.56	15.9093+-0.0329	38.9159+-0.0329	0.9747+-0.0005	0.004189+-0.000073	0.000131+-0.000001	0.011382+-0.000043	0.398803+-0.003495	7.9922+-0.0759	0.143179+-0.001064	16.8937+-0.0645	1.973087+-0.099471	-5.8083+-0.4441	0.804399+-0.021533	1.9187+-0.2312	0.243824+-0.000000	28.75-41.8	0.149839+-0.001259	16.5090+-0.0723	0.023417+-0.000000	7-15.7	0.312591+-0.005099	10.1085+-0.1418	0.619438+-0.000000	15.7-28.75	0.143503+-0.000822	16.8827+-0.0485	0.006539+-0.000000	41.8-50.5	0.337944+-0.004222	9.4342+-0.1109
ResUNet	8.11	13.0502+-0.0999	36.0568+-0.0999	0.9630+-0.0007	0.005982+-0.000151	0.000253+-0.000006	0.015824+-0.000182	0.580493+-0.004514	4.7277+-0.0675	0.202740+-0.001825	13.8691+-0.0791	2.196548+-0.210000	-6.7265+-0.8433	0.993230+-0.044053	0.0782+-0.3850	0.243824+-0.000000	28.75-41.8	0.203625+-0.003579	13.8350+-0.1530	0.023417+-0.000000	7-15.7	0.323164+-0.002858	9.8146+-0.0767	0.619438+-0.000000	15.7-28.75	0.215460+-0.001770	13.3486+-0.0718	0.006539+-0.000000	41.8-50.5	0.331564+-0.004233	9.6056+-0.1129
Attention UNet	7.85	13.4453+-0.1919	36.4519+-0.1918	0.9651+-0.0010	0.005995+-0.000201	0.000230+-0.000010	0.015111+-0.000333	0.550504+-0.008220	5.1888+-0.1297	0.194943+-0.004859	14.2127+-0.2146	1.883427+-0.132940	-5.4135+-0.5909	0.968391+-0.028346	0.2907+-0.2566	0.243824+-0.000000	28.75-41.8	0.196462+-0.003245	14.1463+-0.1433	0.023417+-0.000000	7-15.7	0.317140+-0.003001	9.9775+-0.0823	0.619438+-0.000000	15.7-28.75	0.200722+-0.003868	13.9657+-0.1676	0.006539+-0.000000	41.8-50.5	0.376004+-0.009141	8.5124+-0.2100

Dataset Variant T02_mod

Method	Parameters (M)	SNR	PSNR	SSIM	MAE	MSE	RMSE	EB_WSE_MEDIUM_40_70_NE	EB_WSE_MEDIUM_40_70_SNR	EB_WSE_STRONG_70_100_NE	EB_WSE_STRONG_70_100_SNR	EB_WSE_VERY_WEAK_5_20_NE	EB_WSE_VERY_WEAK_5_20_SNR	EB_WSE_WEAK_20_40_NE	EB_WSE_WEAK_20_40_SNR	FB_FRE_HIGH_ENERGY_RATIO	FB_FRE_HIGH_FREQUENCY_RANGE_HZ	FB_FRE_HIGH_NE	FB_FRE_HIGH_SNR	FB_FRE_LOW_ENERGY_RATIO	FB_FRE_LOW_FREQUENCY_RANGE_HZ	FB_FRE_LOW_NE	FB_FRE_LOW_SNR	FB_FRE_MID_ENERGY_RATIO	FB_FRE_MID_FREQUENCY_RANGE_HZ	FB_FRE_MID_NE	FB_FRE_MID_SNR	FB_FRE_VERY_HIGH_ENERGY_RATIO	FB_FRE_VERY_HIGH_FREQUENCY_RANGE_HZ	FB_FRE_VERY_HIGH_NE	FB_FRE_VERY_HIGH_SNR
Input	-	3.3315+-0.0000	26.6881+-0.0000	0.7654+-0.0000	0.012160+-0.000000	0.002216+-0.000000	0.046691+-0.000000	1.812672+-0.000000	-5.0296+-0.0000	0.291407+-0.000000	10.7119+-0.0000	32.195003+-0.000000	-30.0783+-0.0000	10.154543+-0.000000	-19.9091+-0.0000	0.243824+-0.000000	28.75-41.8	0.683361+-0.000000	3.3090+-0.0000	0.023417+-0.000000	7-15.7	0.695802+-0.000000	3.1510+-0.0000	0.619438+-0.000000	15.7-28.75	0.678645+-0.000000	3.3687+-0.0000	0.006539+-0.000000	41.8-50.5	0.689609+-0.000000	3.2288+-0.0000
UNet	7.76	14.7039+-0.2431	38.0605+-0.2431	0.9748+-0.0005	0.004650+-0.000281	0.000164+-0.000009	0.012648+-0.000360	0.473852+-0.018317	6.5041+-0.3309	0.167761+-0.005444	15.5276+-0.2822	2.123491+-0.119411	-6.5083+-0.4739	0.907897+-0.009668	0.8630+-0.0887	0.243824+-0.000000	28.75-41.8	0.168928+-0.004161	15.4765+-0.2150	0.023417+-0.000000	7-15.7	0.270377+-0.001948	11.3674+-0.0602	0.619438+-0.000000	15.7-28.75	0.177198+-0.010899	15.0684+-0.5435	0.006539+-0.000000	41.8-50.5	0.314485+-0.026862	10.0991+-0.7195
DnCNN	0.56	17.4002+-0.0460	40.7567+-0.0460	0.9847+-0.0002	0.002816+-0.000052	0.000088+-0.000001	0.009280+-0.000054	0.317468+-0.002446	9.9740+-0.0672	0.123480+-0.000698	18.1918+-0.0488	1.908495+-0.075431	-5.5801+-0.3437	0.699431+-0.018116	3.1511+-0.2222	0.243824+-0.000000	28.75-41.8	0.127505+-0.000947	17.9398+-0.0564	0.023417+-0.000000	7-15.7	0.268126+-0.002124	11.4418+-0.0681	0.619438+-0.000000	15.7-28.75	0.119451+-0.000877	18.4804+-0.0641	0.006539+-0.000000	41.8-50.5	0.295901+-0.003349	10.5996+-0.0971
ResUNet	8.11	14.1138+-0.0794	37.4704+-0.0794	0.9749+-0.0005	0.004629+-0.000126	0.000187+-0.000004	0.013524+-0.000125	0.486774+-0.004043	6.2610+-0.0725	0.181058+-0.001342	14.8585+-0.0644	2.130174+-0.187739	-6.5089+-0.7907	0.872875+-0.023638	1.2102+-0.2271	0.243824+-0.000000	28.75-41.8	0.180487+-0.003511	14.8949+-0.1682	0.023417+-0.000000	7-15.7	0.270266+-0.001293	11.3680+-0.0415	0.619438+-0.000000	15.7-28.75	0.192449+-0.001287	14.3336+-0.0586	0.006539+-0.000000	41.8-50.5	0.290352+-0.006327	10.7610+-0.1905
Attention UNet	7.85	14.5616+-0.1395	37.9182+-0.1395	0.9749+-0.0004	0.004868+-0.000111	0.000168+-0.000005	0.012843+-0.000208	0.455199+-0.003818	6.8466+-0.0745	0.172010+-0.003499	15.3022+-0.1770	2.171944+-0.038920	-6.7162+-0.1614	0.884095+-0.010330	1.1019+-0.0963	0.243824+-0.000000	28.75-41.8	0.174245+-0.003430	15.1990+-0.1717	0.023417+-0.000000	7-15.7	0.270073+-0.005210	11.3789+-0.1665	0.619438+-0.000000	15.7-28.75	0.176185+-0.003015	15.0971+-0.1458	0.006539+-0.000000	41.8-50.5	0.334154+-0.012100	9.5456+-0.3116

Dataset Variant T02_simp

Method	Parameters (M)	SNR	PSNR	SSIM	MAE	MSE	RMSE	EB_WSE_MEDIUM_40_70_NE	EB_WSE_MEDIUM_40_70_SNR	EB_WSE_STRONG_70_100_NE	EB_WSE_STRONG_70_100_SNR	EB_WSE_VERY_WEAK_5_20_NE	EB_WSE_VERY_WEAK_5_20_SNR	EB_WSE_WEAK_20_40_NE	EB_WSE_WEAK_20_40_SNR	FB_FRE_HIGH_ENERGY_RATIO	FB_FRE_HIGH_FREQUENCY_RANGE_HZ	FB_FRE_HIGH_NE	FB_FRE_HIGH_SNR	FB_FRE_LOW_ENERGY_RATIO	FB_FRE_LOW_FREQUENCY_RANGE_HZ	FB_FRE_LOW_NE	FB_FRE_LOW_SNR	FB_FRE_MID_ENERGY_RATIO	FB_FRE_MID_FREQUENCY_RANGE_HZ	FB_FRE_MID_NE	FB_FRE_MID_SNR	FB_FRE_VERY_HIGH_ENERGY_RATIO	FB_FRE_VERY_HIGH_FREQUENCY_RANGE_HZ	FB_FRE_VERY_HIGH_NE	FB_FRE_VERY_HIGH_SNR
Input	-	6.6896+-0.0000	29.5209+-0.0000	0.8797+-0.0000	0.006035+-0.000000	0.001162+-0.000000	0.033769+-0.000000	1.359040+-0.000000	-2.5295+-0.0000	0.218058+-0.000000	13.2445+-0.0000	18.050747+-0.000000	-24.8266+-0.0000	5.957535+-0.000000	-15.2400+-0.0000	0.243824+-0.000000	28.75-41.8	0.465783+-0.000000	6.6411+-0.0000	0.023417+-0.000000	7-15.7	0.478424+-0.000000	6.4075+-0.0000	0.619438+-0.000000	15.7-28.75	0.461325+-0.000000	6.7222+-0.0000	0.006539+-0.000000	41.8-50.5	0.470025+-0.000000	6.5676+-0.0000
UNet	7.76	17.0387+-0.2113	39.8701+-0.2112	0.9844+-0.0008	0.003486+-0.000332	0.000108+-0.000005	0.010272+-0.000247	0.365571+-0.003555	8.7478+-0.0849	0.127483+-0.003382	17.9170+-0.2360	1.415779+-0.097544	-2.9125+-0.5822	0.652592+-0.002307	3.7166+-0.0316	0.243824+-0.000000	28.75-41.8	0.124367+-0.002468	18.1415+-0.1829	0.023417+-0.000000	7-15.7	0.215077+-0.010455	13.3626+-0.4145	0.619438+-0.000000	15.7-28.75	0.133467+-0.001234	17.5222+-0.0794	0.006539+-0.000000	41.8-50.5	0.251275+-0.022851	12.0500+-0.7735
DnCNN	0.56	20.4091+-0.0326	43.2404+-0.0326	0.9925+-0.0000	0.001532+-0.000031	0.000049+-0.000000	0.006957+-0.000029	0.222201+-0.001296	13.0774+-0.0526	0.088844+-0.000339	21.0477+-0.0335	0.898625+-0.021335	1.0318+-0.2143	0.406727+-0.002426	7.8283+-0.0523	0.243824+-0.000000	28.75-41.8	0.086655+-0.000918	21.2753+-0.0958	0.023417+-0.000000	7-15.7	0.188400+-0.003696	14.5150+-0.1741	0.619438+-0.000000	15.7-28.75	0.086650+-0.000717	21.2743+-0.0735	0.006539+-0.000000	41.8-50.5	0.202190+-0.000115	13.8956+-0.0046
ResUNet	8.11	16.5535+-0.1634	39.3848+-0.1634	0.9847+-0.0004	0.003306+-0.000134	0.000121+-0.000004	0.010861+-0.000203	0.382057+-0.009326	8.3679+-0.2092	0.136025+-0.002203	17.3502+-0.1454	1.356662+-0.063744	-2.5668+-0.3877	0.618934+-0.018298	4.1788+-0.2558	0.243824+-0.000000	28.75-41.8	0.131448+-0.003584	17.6564+-0.2426	0.023417+-0.000000	7-15.7	0.211059+-0.001684	13.5171+-0.0696	0.619438+-0.000000	15.7-28.75	0.144846+-0.002013	16.8097+-0.1269	0.006539+-0.000000	41.8-50.5	0.222782+-0.007011	13.0636+-0.2739
Attention UNet	7.85	16.6680+-0.1592	39.4993+-0.1592	0.9831+-0.0005	0.003818+-0.000183	0.000117+-0.000004	0.010712+-0.000193	0.345422+-0.001738	9.2426+-0.0445	0.132686+-0.002938	17.5634+-0.1929	1.558382+-0.103877	-3.7426+-0.5584	0.655009+-0.031865	3.7075+-0.4077	0.243824+-0.000000	28.75-41.8	0.129747+-0.002258	17.7680+-0.1561	0.023417+-0.000000	7-15.7	0.227748+-0.006546	12.8649+-0.2507	0.619438+-0.000000	15.7-28.75	0.136028+-0.000966	17.3499+-0.0621	0.006539+-0.000000	41.8-50.5	0.282750+-0.011078	10.9940+-0.3458