LAVA-WAAM: A Dataset for Segmentation and Tracking of Material Deposition in Wire Arc Additive Manufacturing

LAVA-WAAM is a dataset for segmentation and tracking of material deposition in Wire Arc Additive Manufacturing. The name LAVA stands for Layer Analysis and Visual Assessment, highlighting its focus on tracking and control of deposited layers during the build process. The dataset captures challenging segmentation scenarios characterised by a wide range of variabilities and disturbances, including: Diverse process conditions and anomalies Dynamic plasma illumination Variations in build geometry Non-uniform deposition widths Material deformations Multiple workpiece perspectives Varying camera poses More details about the dataset are described in the corresponding paper. The hosted dataset contains images, annotations, and related files. Please visit our GitHub page for more information about the source code and models. Data Structure The LAVA-WAAM dataset follows the style of the DISK5K dataset, with a similar folder hierarchy. The dataset is primarily divided into 15 deposition sessions. Each session is further divided into subsets. Each subset directory, for example E1_S1_W1_L2, contains images and masks (ground truth) folders. . ├── E1_S1_W1 │ ├── E1_S1_W1_L1 │ ├── E1_S1_W1_L2 │ ├── images │ ├── masks │ ... │ ├── E2_S2_F1 ├── E2_S3_W1 ├── E3_S4_C1 ├── E3_S5_C2 ├── E4_S6_C0 ├── E5_S7_C0 ├── E5_S8_C0 ├── E5_S9_C1 ├── E5_S10_C0 ├── E5_S11_C2 ├── E6_S12_T1 ├── E6_S13_C1 ├── E6_S14_C2 └── E6_S15_C3 Naming Convention We use the following nomenclature to group collected data in an intuitive and descriptive way: EX_SY_SHAPE_Z. EX - the number of the experiment block (1 to 6), SY - the number of the deposition session (1 to 15), SHAPE - the shape of the build (W – wall, F – flat, C – cylinder, T – trophy-shaped) Z - the build number (0 indicates a trial build) For more details please refer to Table V and Table VI in our publication. Citation The LAVA-WAAM dataset was released as part of following work: @article{..., author = {...}, title = {...}, journal = {...}, year = {...}, ... }