## 700 annotations of Kumar Sangakkara's face ## **Context** Recently I have been working on some object localization problems using Convolutional Nets and I wanted to try train the model on a new dataset other than the very common COCO or PASCAL VOC datasets. While pondering on what object to compile a small dataset around, I thought of pushing the challenge a bit more to see if the same model can be trained to localize faces. Having this in mind I wanted a dataset of a person's face annotations. As you may know with Deep Learning models, the more data you have the more accuracy you reach. So considering the challenge to detect a face I wanted a considerable number of images of the same face that the model should be trained on. Hence, I needed many pictures of the same person. So the person had to be famous so I could easily find many pictures. So being in Sri Lanka where else to look other than our Cricket stars. So I chose the living legend in Sri Lankan Cricket, Kumar Sangakkara. **Content** I downloaded around 1000 images from google images and after manual cleaning ended up with 704, which are contained here. I manually annotated all the pictures using a python script to generate the xml files. (Yeah, I couldn't find a better thing to do in that 2 hours.) Now here is the dataset for anyone to make use of. The zip file attached contains two folders, images and annotations. **Inspiration** So as I mentioned in the above description, my goal with this dataset was to see if an object localization model can be used to detect a face of a person. Even though I have the pipeline, I couldn't still thoroughly test its performance using a GPU. So anyone whose interested can use this dataset to test those results. Also if these annotations can be useful for any other application, feel free to use it and share it. Have fun!