Skip to content

Releases: AI4EPS/models

PhaseNet-Plus-XUNet-v0

06 Aug 15:44
6dc992b

Choose a tag to compare

Create LICENSE

PhaseNet-Plus-UNet-v1

06 Aug 15:38
6dc992b

Choose a tag to compare

Create LICENSE

PhaseNet-Plus-UNet-v0

04 Jul 23:13
6dc992b

Choose a tag to compare

PhaseNet+ with new U-Net backbone

PhaseNet-DAS-v1

12 Jul 23:54
6dc992b

Choose a tag to compare

Create LICENSE

PhaseNet-DAS-v0

21 Jul 20:31
6dc992b

Choose a tag to compare

Create LICENSE

PhaseNet-DAS-Plus-Arcata-v1

24 Mar 03:24
6dc992b

Choose a tag to compare

PhaseNet-DAS-Plus (Arcata)

Semi-supervised DAS phase picker trained on Arcata (Northern California) DAS data.

Model Details

  • Architecture: PhaseNet-DAS-Plus (UNet backbone with event detection heads)
  • Training data: 136 labeled events from Arcata DAS array, labels generated by PhaseNet-2018 with PS-pair association (origin_window=0.1, moveout_window=0.2)
  • Input: (1, nx, nt) DAS strain rate, nx=2048, nt=4096
  • Output: Phase picks (P/S/Noise) + event center + event time
  • Training: 50k iterations, lr=1e-4, EMA decay=0.999, batch_size=2, num_patch=16
  • Parameters: 9.3M

Usage

import torch
import phasenet.models.phasenet_das_plus as model_module

ckpt = torch.load("phasenet_das_plus_arcata_v1.pth", map_location="cpu")
model = model_module.build_model(**ckpt["model_config"])
model.load_state_dict(ckpt["model_ema_weights"])  # Use EMA weights
model.eval()

DASNet-v1

07 Apr 06:33
6dc992b

Choose a tag to compare

DASNet v1

Mask R-CNN for event detection and classification in Distributed Acoustic Sensing (DAS) data from submarine cables.

Model Details

  • Architecture: Mask R-CNN with ResNet50-FPN backbone (selectable P4 layer)
  • Training data: SeaFOAM Monterey Bay DAS deployment (Romanowicz et al., 2023)
  • Input: (1, nx, nt) DAS strain rate, preprocessed to 3-channel RGB (raw, 2–10 Hz bandpass, >10 Hz highpass)
  • Output: Bounding boxes, class labels, confidence scores, instance masks
  • Classes: Blue whale A/B/D, Fin whale, P wave, S wave, T wave, Ship, Others (10 classes + background)
  • Parameters: ~44M

Usage

import torch
from dasnet import build_dasnet_model, load_checkpoint

model = build_dasnet_model()
load_checkpoint(model, "dasnet_v1.pth", torch.device("cpu"))

Reference

Zhang, C., et al. (2026). "A deep learning framework for marine acoustic and seismic monitoring with distributed acoustic sensing." arXiv:2603.14844

PhaseNet-v1

12 May 20:18
6dc992b

Choose a tag to compare

PhaseNet-v1 Pre-release
Pre-release
Create LICENSE

PhaseNet-TF-LCSN

24 Apr 22:31
6dc992b

Choose a tag to compare

PhaseNet-TF-LCSN Pre-release
Pre-release
Create LICENSE

PhaseNet-Plus-v1

01 Jan 06:46
6dc992b

Choose a tag to compare

PhaseNet-Plus-v1 Pre-release
Pre-release
Create LICENSE