Combined Vessel Dataset (COVD) Results

F1 Scores (micro-level)

• Benchmark results for models: DRIU, HED, M2U-Net and U-Net.

• Models are trained on a COVD excluding the target dataset, and tested on the target dataset (numbers in bold indicate number of parameters per model). Models are trained for a fixed number of 1000 epochs, with a learning rate of 0.001 until epoch 900 and then 0.0001 until the end of the training.

• Database and model resource configuration links (table top row and left column) are linked to the originating configuration files used to obtain these results.

• Check our paper for details on the calculation of the F1 Score and standard deviations (in parentheses).

• Single performance numbers correspond to a priori performance indicators, where the threshold is previously selected on the training set (COVD excluding the target dataset)

• You can cross check the analysis numbers provided in this table by downloading this software package, the raw data, and running bob binseg analyze providing the model URL as --weight parameter.

• For comparison purposes, we provide “second-annotator” performances on the same test set, where available.

		driu	hed	m2unet	unet
Dataset	2nd. Annot.	15M	14.7M	0.55M	25.8M
drive	0.788 (0.021)	0.768 (0.031)	0.750 (0.036)	0.771 (0.027)	0.775 (0.029)
stare	0.759 (0.028)	0.786 (0.100)	0.738 (0.193)	0.800 (0.080)	0.806 (0.072)
chasedb1	0.768 (0.023)	0.778 (0.031)	0.777 (0.028)	0.776 (0.031)	0.779 (0.028)
hrf		0.742 (0.049)	0.719 (0.047)	0.735 (0.045)	0.746 (0.046)
iostar-vessel		0.790 (0.023)	0.792 (0.020)	0.788 (0.021)	0.783 (0.019)