|
|
Counting of characteristic scales of sand lizards in colour images
Maršala, Štěpán ; Štursa, Dominik (referee) ; Škrabánek, Pavel (advisor)
The diploma thesis describes the design and implementation of a program for counting secondary scales in the image data of the ventral sides of the bodies of sand lizards. The program respects the requirements of scientists from the Institute of Vertebrate Biology of the Czech Academy of Sciences and the Faculty of Education at Masaryk University for the controllability and accuracy of results. The program consists of several parts. In input receives photos of sand lizards, in which he cuts out an area of interest. Unifies the orientation of these sections using detected objects. Object detection is provided by YOLOv4. Another part of the program called the Centroid Detector determines the position of the centers of the secondary scales in the unified sections. This part uses the U-Net convolutional neural network, which is specially modified to detect the centers of objects in close proximity. The other parts of the program divide the detected positions of the scale centers into left and right secondary rows and write their numbers to the output file.
|
|
|
Adaptation of a centroid-based object detector on multiclass tasks
Kelbl, Jan ; Dobrovský, Ladislav (referee) ; Škrabánek, Pavel (advisor)
Tato diplomová práce demonstruje využitelnost centroid detektoru na úlohách vícetřídní detekce objektů. Teoretická část této práce se zaměřuje na základy digitálního obrazu, umělé neuronové sítě, techniky segmentace obrazu s ohledem na vícetřídní úlohy a techniky detekce objektů, obojí s využitím hlubokého učení. Praktická část se zaměřuje na adaptaci centroid detektoru na vícetřídní úlohy včetně návrhu techniky tréninku. Je zavedena nová metrika, střední lokalizační odchylka, umožňující nezkreslené vyhodnocení vícetřídního centroid detektoru. Nakonec je vícetřídní centroid detektor natrénován a vyhodnocen na třech datových sadách různé složitosti. Výsledky ukazují skvělou výkonnost centroid detektoru na vícetřídních úlohách. To je důležitým krokem k aplikaci centroid detektoru na složitější reálné úlohy.
|
|
|
Counting of characteristic scales of sand lizards in colour images
Maršala, Štěpán ; Štursa, Dominik (referee) ; Škrabánek, Pavel (advisor)
The diploma thesis describes the design and implementation of a program for counting secondary scales in the image data of the ventral sides of the bodies of sand lizards. The program respects the requirements of scientists from the Institute of Vertebrate Biology of the Czech Academy of Sciences and the Faculty of Education at Masaryk University for the controllability and accuracy of results. The program consists of several parts. In input receives photos of sand lizards, in which he cuts out an area of interest. Unifies the orientation of these sections using detected objects. Object detection is provided by YOLOv4. Another part of the program called the Centroid Detector determines the position of the centers of the secondary scales in the unified sections. This part uses the U-Net convolutional neural network, which is specially modified to detect the centers of objects in close proximity. The other parts of the program divide the detected positions of the scale centers into left and right secondary rows and write their numbers to the output file.
|
guest ::
