|
|
@@ -17,7 +17,7 @@ folder.
|
|
|
We also provide the annotations from our algorithm as detected from the
|
|
|
[sp\_tool](https://github.com/MikhailStartsev/sp_tool), which is based on
|
|
|
Agtzidis et al. (2016), along with annotations from Berg et al. (2009) and Larsson et al.
|
|
|
-(2015) in the corresponding folders starting with *output_*.
|
|
|
+(2015) (our re-implementation of the latter is available on the http://michaeldorr.de/smoothpursuit/ page, or via this [link](http://michaeldorr.de/smoothpursuit/larsson_reimplementation.zip)) in the corresponding folders starting with *output_*.
|
|
|
|
|
|
Finally, a collection of 45 hand annotated targets is provided in the
|
|
|
[targets\_arff](https://web.gin.g-node.org/ioannis.agtzidis/gazecom_annotations/src/master/targets_arff)
|
|
|
@@ -25,10 +25,17 @@ folder.
|
|
|
|
|
|
# sp_tool
|
|
|
|
|
|
-The implementation of our detection algorithm together with a wide variety of
|
|
|
+The implementation of our clustering-based detection algorithm together with a wide variety of
|
|
|
evaluation metrics for eye movement classification can be found
|
|
|
[here](https://github.com/MikhailStartsev/sp_tool).
|
|
|
|
|
|
+# Model performance comparison
|
|
|
+A tentative evaluation table of sample-level model evaluation statistics for eye movement classification on GazeCom is available at http://michaeldorr.de/smoothpursuit/.
|
|
|
+
|
|
|
+For a more thorough evaluation (all eye movement classes; including event-level metrcis), plese refer to our later papers:
|
|
|
+* [1D CNN with BLSTM for automated classification of fixations, saccades, and smooth pursuits](https://link.springer.com/article/10.3758/s13428-018-1144-2) (Startsev et al., 2019, Behavior Research Methods).
|
|
|
+* [A novel gaze event detection metric that is not fooled by gaze-independent baselines](https://dl.acm.org/citation.cfm?id=3319836) (Startsev et al., 2019, ETRA)
|
|
|
+
|
|
|
## References
|
|
|
|
|
|
Agtzidis, I., Startsev, M., & Dorr, M. (2016). Smooth pursuit detection
|
