Fig. 4: Application of DenseSSD to solution-filled vial datasets. | npj Computational Materials

Fig. 4: Application of DenseSSD to solution-filled vial datasets.

From: Machine vision-based detections of transparent chemical vessels toward the safe automation of material synthesis

Fig. 4

a Compositions of the training datasets used for the detection experiments. The number refers to the number of images in the training dataset. b The appearances of empty vials and solution-filled vials. Image processing techniques based on hue and saturations were applied to the images involving solution-filled vials for data augmentations. c Detection performance (AP) comparisons between YOLOv3, DETR, SSD, YOLOv6, and DenseSSD under different training set construction procedures: one that conducts training with only the empty vial dataset and another that conducts training with the dataset containing both empty and filled vials. d Exemplary detection results with other benchmark models and DenseSSD, achieved for two scenes involving multiple solution-filled vials. One scene (Case 1) contains four vials including one success cases and three failure cases with the lean-in and stand-on statuses, while the other one (Case 2) contains seven vials including four success cases and three failure cases with the stand-on and lean-in statuses. The failure vial cases are highlighted in yellow dotted lines for clarity.

Back to article page