Computer vision postgraduate personal statement example

Personal statement example

My final-year project classified photographs of household packaging. The model seemed to work well until I photographed the same objects on a different kitchen table. Several confident predictions changed, even though the packaging had not. Looking at those mistakes was more interesting than improving the original score: I wanted to understand which parts of an image the system was using and why its decisions were so sensitive to the setting. That question is my main reason for applying for postgraduate study in computer vision.

During my computing degree I enjoyed the combination of programming and mathematical reasoning in modules on algorithms, linear algebra and probability. I became comfortable implementing methods in Python, but I also learned to work through small examples by hand before trusting a library call. Matrix operations stopped being abstract exercises when I used them to represent image transformations. Probability made me more cautious about treating a model's most likely answer as a reliable one.

For my project I collected a small dataset using my phone, with permission to photograph packaging brought in by family members. I compared a simple classifier based on colour and edge features with a small convolutional network. Initially I split individual photographs at random. My supervisor pointed out that several photographs of the same object could therefore appear in both training and test sets. I rebuilt the split around the physical objects, keeping all views of each object together, and reported the less flattering result. It was an important lesson in evaluating the task I actually wanted to solve rather than the dataset I happened to have.

I then kept a record of errors by lighting, background and object type. Transparent containers and damaged labels were particularly difficult, and some categories were ambiguous even to me. Instead of quietly discarding these cases, I described the labelling decisions and the limits they placed on my conclusions. The project remained a classroom investigation with a small collection of objects; it did not establish that the system would work in a recycling facility. I would like to develop stronger experimental methods and a better understanding of representation learning so that I can investigate such problems with more rigour.

Alongside university I worked two evenings a week in a supermarket. Closing the checkout area and reconciling the till required care when I was tired, and arranging shifts around project deadlines taught me to plan work realistically. I also help at a monthly community repair session, mainly setting up donated laptops and explaining basic settings under an experienced volunteer's guidance. These conversations remind me that a technically correct explanation is not always a useful one. I have become better at asking what someone is trying to do before suggesting a solution.

My next step is to move beyond training an existing network and understand more of the reasoning behind its design and evaluation. I am interested in how visual systems handle changes in viewpoint, incomplete information and settings that differ from their training data. I would welcome advanced study in image geometry, learning methods and experimental design, together with a substantial project in which I can test a focused question. In the longer term I hope to work on visual inspection or image-analysis software where reporting uncertainty matters as much as producing a prediction. I bring practical programming experience, a willingness to revise an experiment and the habit of taking an inconvenient result seriously.