[WEEK 3-Facade Parsing using Deep Learning]
Team Members: furkan karababa, Onur Cankur , Javid Rajabov
Last week, we looked for source code for sample model of our project, so we asked for source code from authors of related papers of our project via email. However they didn’t reply to our emails. So, we decided to look for source code elsewhere.
Labeling Dataset
This week, we learned the way how to labeling our image dataset. There is one tool for this usage named labelme, here is the link for github address of that tool. I the future steps our project, if we need more labelled dataset, we will label our images by using this tool in the coming weeks.
Semantic Segmentation Methods and Models
We are going to use semantic segmentation in image parsing step of our project. In the first place, the main idea in segmentation is that images are input of it, and output is regions in the form of line segments and curve segments. Secondly, the process of understanding, recognizing things in the image in pixel level is called semantic segmentation.
This week, we investigated well-known methods of semantic segmentation. And here is the link for wide explanation of semantic segmentation which we benefited from while learning it. There are 3 main approaches for semantic segmentation:
1 — Region-Based Semantic Segmentation
The region-based methods generally follow the “segmentation using recognition” pipeline, which first extracts free-form regions from an image and describes them, followed by region-based classification.
2 — Fully Convolutional Network-Based Semantic Segmentation
The original Fully Convolutional Network learns a mapping from pixels to pixels, without extracting the region proposals. The FCN network pipeline is an extension of the classical CNN. The main idea is to make the classical CNN take as input arbitrary-sized images.
3 — Weakly Supervised Semantic Segmentation
Most of the relevant methods in semantic segmentation rely on a large number of images with pixel-wise segmentation masks. However, manually annotating these masks is quite time-consuming, frustrating and commercially expensive. Therefore, some weakly supervised methods have recently been proposed, which are dedicated to fulfilling the semantic segmentation by utilizing annotated bounding boxes.
Deciding and Running one of the model
We are more interested in Fully Convolutional Network, we are researching it widely. We found github for semantic segmentation which has many models and implementations for it such as Encoder_Decoder, DeepLab, SegNet and others. And we are going to run source code according to these models separately, then compare the results of them in order to choose baseline models. Briefly, we are working on running these models and decide to which of them we will use and according to these we will write our project progress report.
