Computer vision for recognition segmentation and classification of materials and vessels in the chemistry lab (and other settings)
The methods to in this repository are aimed to detect segment and classify vessels, and material phases, in mostly transparent containers/vessels in the chemistry lab. All the methods here are based on convolutional neural nets (CNN). The methods were trained using the Vector-LabPics dataset.
Network for semantic segmentation of both vessels and materials using fully convolutional net can be found here:
This basically assigns one or more class per pixel. The classes includes vessel or various of materials class such as liquids,solids powder,foam... (Figure 1).
Nets that split the image to instances of materials and vessels (Figure 2) can be found here:
Generator evaluator selector net (Hierarchical)
and:
The Vector LabPics dataset contains images of materials and vessels in the chemistry lab and other settings. The images are annotated for both the materials and vessels for both semantic and instance segmentation. Examples can be seen in Figures 1 and 2. The full dataset can be download from Here or here
Figure 1 : Examples for semantic segmentation from the LabPics dataset (GT) and net results (Pred)(Images taken from the NileRed youtube channel).
Figure 1: Examples for instance segmentation from the LabPics dataset (GT) and net results (Pred)(Images taken from the NileRed youtube channel).
The images for the LabPics dataset were supplied by the following sources Nessa Carson (@SuperScienceGrl Twitter), Chemical and Engineering Science chemistry in pictures, YouTube channels dedicated to chemistry experiments: NurdRage, NileRed, DougsLab, ChemPlayer, and Koen2All. Additional sources for images include Instagram channels chemistrylover_(Joana Kulizic),Chemistry.shz (Dr.Shakerizadeh-shirazi), MinistryOfChemistry, Chemistry And Me, ChemistryLifeStyle, vacuum_distillation, and Organic_Chemistry_Lab