|
|
Machine Learning on Synthetic Data for Counting Crates in Images
Koďousek, Ondřej ; Juránek, Roman (referee) ; Herout, Adam (advisor)
The goal of this work is to create a process that counts how many crates are in a video or still image. This is done by using a model that is trained on a synthetic dataset, and then the results are adjusted at the individual frame level and then at the continuous video frame level. This synthetic dataset is generated using a script in Blender using Octane Render, for a higher level of photorealism. The benefit of successfully training on the synthetic dataset is faster and especially automatic annotation. Since the annotations are generated with the image itself, it is not a problem to generate a large number of images without a single manual annotation. Another benefit is a head start in model generation for detecting objects that are new to the market and lack sufficient data, or are only in production. I have detected in still images and video, and in both cases I achieved success rates above 90% with a model trained on synthetic data.
|
|
|
Detection of Graffiti Tags in Image
Molisch, Marek ; Herout, Adam (referee) ; Špaňhel, Jakub (advisor)
The goal of this work is to compare today's architecture of object detection models and use them for the purpose of graffiti tag detection. State-of-the-art models, which are compatible with the Tensorflow framework, were used. Faster R-CNN architecture was found to be the most accurate and SSD architecture to be the fastest. Experiments with graffiti tags from Athens in the STORM dasater showed, that it is better to approach graffiti tags as objects rather than writings.
|
|
|
Detection of Graffiti Tags in Image
Molisch, Marek ; Herout, Adam (referee) ; Špaňhel, Jakub (advisor)
The goal of this work is to compare today's architecture of object detection models and use them for the purpose of graffiti tag detection. State-of-the-art models, which are compatible with the Tensorflow framework, were used. Faster R-CNN architecture was found to be the most accurate and SSD architecture to be the fastest. Experiments with graffiti tags from Athens in the STORM dasater showed, that it is better to approach graffiti tags as objects rather than writings.
|
guest ::
