The current study examined how repetitive exposure of an opponent’s stroke preferences on video affected the integration of contextual priors and kinematic information during anticipation in skilled boxers. We performed an experimental ingroup-design with a temporal-occlusion video-based anticipation task with repeated measures where 19 male skilled boxers (M = 22.95 age, SD = 4.26) classified as A boxers in Sweden participated in the study. The test represented three different stroke combinations divided into four tests and two exposure videos. Each test included 22 occluded clips divided into two blocks and each block contained six high-probability strokes, three moderate-probability strokes and two low- probability strokes. The participants were informed to answer which stroke they anticipated and how sure they were of their answer after each clip. Retrospective verbal reports were answered regarding what information the participants used to anticipate the strokes in the preceding test. The purpose of the exposure videos was to manipulate the participants’ to-be- anticipated action in favor of the opponent’s stroke preferences both when the opponent acted and did not act accordingly. The result indicated that participants learned kinematic information about the opponent by observing the opponent on video rather than learning about the opponent’s stroke preferences. Future research is needed to fully understand how contextual priors integrate with kinematic information in boxing. / Föreliggande studie undersökte hur repetitiv exponering av en motståndares slagpreferenser på video påverkade integrationen av kontextuell förhandsinformation och kinematisk information under antecipering hos skickliga boxare. Vi utförde en experimentell inomgruppsdesign med en videobaserad temporal-ocklusion anticiperingsuppgift med upprepade mätningar där 19 skickliga manliga boxare (M = 22.95 ålder, SD = 4.26) klassade som A-boxare i Sverige deltog i studien. Testet representerade tre olika slagkombinationer uppdelade i fyra tester och två exponerings videor. Varje test inkluderade 22 ockluderade klipp uppdelade i två block där varje block innehöll sex slag med hög sannolikhet, tre slag med måttlig sannolikhet och två slag med låg sannolikhet. Deltagarna informerades om att svara vilket slag de anteciperade och hur säkra de var på sitt svar efter varje klipp. Retrospektiva verbala rapporter besvarades angående vilken information deltagarna använde för att antecipera slagen i det föregående testen. Syftet med exponerings videorna var att manipulera deltagarnas antecipering till förmån för motståndarens slag preferenser både när motståndaren agerade och inte agerade därefter. Resultatet visade att deltagarna lärde sig kinematisk information om motståndaren genom att observera motståndaren på video snarare än att lära sig om motståndarens slag preferenser. Framtida forskning behövs för att öka förståelsen för hur kontextuell förhandsinformation integreras med kinematisk information i boxning.

Using computer vision to automatically process and understand images is becoming increasingly popular. One frequently used technique in this area is object detection, where the goal is to both localize and classify objects in images. Today's detection models are accurate, but there is still room for improvement. Most models process objects independently and do not take any contextual information into account in the classification step. This thesis will therefore investigate if a performance improvement can be achieved by classifying all objects jointly with the use of contextual information. An architecture that has the ability to learn relationships of this type of information is the transformer. To investigate what performance that can be achieved, a new architecture is constructed where the classification step is replaced by a transformer block. The model is trained and evaluated on document images and shows promising results with a mAP score of 87.29. This value is compared to a mAP of 88.19, which was achieved by the object detector, Mask R-CNN, that the new model is built upon.  Although the proposed model did not improve the performance, it comes with some benefits worth exploring further. By using contextual information the proposed model can eliminate the need for Non-Maximum Suppression, which can be seen as a benefit since it removes one hand-crafted process. Another benefit is that the model tends to learn relatively quickly and a single pass over the dataset seems sufficient. The model, however, comes with some drawbacks, including a longer inference time due to the increase in model parameters. The model predictions are also less secure than for Mask R-CNN. With some further investigation and optimization, these drawbacks could be reduced and the performance of the model be improved.