Well it's 100 degrees outside today, so I haven't yet made it out there. Last week, my camera phone shut off twice and it was only 90 degrees. Instead, I focused on two related topics.
1. I met with Dr. Mikki McComb-Kobza for an hour at a local tea shop. She is the director of the Ocean First dive shop's 501c3 branch. Her primary interest is in shark research. She also does quite a bit of research with local high school students including taking them into the field to work with sharks and sea turtles. Our discussion was mostly a "first meeting" type event where we got to learn a little bit more about our respective backgrounds and interests. The short of it is I'll be looking forward to working with her on processing some of her shark data. This means making sure the camera model is calibrated accurately and turning the crank on a model that might help automate measuring sharks.
2. I spent quite a bit of time this week learning about TensorFlow and going through examples and mini-courses. I figure the more I understand about building effective models, the better I'll understand the best type of data to collect. The shark data may end up being an excellent candidate for some of the machine vision algorithms, so it could be a 2-for-1 scenario.
In other news, I heard a radio segment talk about how disgusted veterans were that famous beach battle locations are now covered in trash. They considered it disrespectful to the men who died on those beaches. I think this is an interesting facet of trash pollution.
I also watched a mini-documentary about the homeless/drug addiction problem in Seattle. The scenes they used to show how bad the problem is were filled with piles of trash. Sporadically, the city sends crews in to clean it all up. Seems like more evidence that what might be needed is some automated way to do it so it happens on a more regular basis than when it reaches some threshold.
UPDATE: Went out tonight when the air was a little cooler. Lots of brainstorming, trying to figure out how to speed up the process. The "final" decision for now is to just collect images of the trash in place and then collect it and take another picture. Then, at home, I'll sort out the trash onto a green screen with controlled lighting. This will break the close coupling I had been planning where each folder would have all the types of pictures. Now it will be more like two datasets: one in the field and one at home. I probably will not try to combine them in post-processing either. I'll see how this goes and if I can do it this way to speed things up dramatically. I need thousands of images...
1. I met with Dr. Mikki McComb-Kobza for an hour at a local tea shop. She is the director of the Ocean First dive shop's 501c3 branch. Her primary interest is in shark research. She also does quite a bit of research with local high school students including taking them into the field to work with sharks and sea turtles. Our discussion was mostly a "first meeting" type event where we got to learn a little bit more about our respective backgrounds and interests. The short of it is I'll be looking forward to working with her on processing some of her shark data. This means making sure the camera model is calibrated accurately and turning the crank on a model that might help automate measuring sharks.
2. I spent quite a bit of time this week learning about TensorFlow and going through examples and mini-courses. I figure the more I understand about building effective models, the better I'll understand the best type of data to collect. The shark data may end up being an excellent candidate for some of the machine vision algorithms, so it could be a 2-for-1 scenario.
In other news, I heard a radio segment talk about how disgusted veterans were that famous beach battle locations are now covered in trash. They considered it disrespectful to the men who died on those beaches. I think this is an interesting facet of trash pollution.
I also watched a mini-documentary about the homeless/drug addiction problem in Seattle. The scenes they used to show how bad the problem is were filled with piles of trash. Sporadically, the city sends crews in to clean it all up. Seems like more evidence that what might be needed is some automated way to do it so it happens on a more regular basis than when it reaches some threshold.
UPDATE: Went out tonight when the air was a little cooler. Lots of brainstorming, trying to figure out how to speed up the process. The "final" decision for now is to just collect images of the trash in place and then collect it and take another picture. Then, at home, I'll sort out the trash onto a green screen with controlled lighting. This will break the close coupling I had been planning where each folder would have all the types of pictures. Now it will be more like two datasets: one in the field and one at home. I probably will not try to combine them in post-processing either. I'll see how this goes and if I can do it this way to speed things up dramatically. I need thousands of images...
Comments