Robin Liu

Robin Liu

OCR, mechanical

Camera Issues

less than 1 minute read

I got around to testing our webcam, and the results were dissapointing to say the least.

Basically, the orginal idea of using a standard video call type webcam off of amazon is just not possible for any reasonable OCR tasks. The resolution is pathetic, the camera is effectively unusable for reading text, and new orders are already on the way. With no prior photography experience on the team, and a limited budget, this was not entirely unexpected. We just should have caught this issue sooner.

For now I am going forward with using smartphone cameras until a better solution is procured. The camera mount attachment on the stand is getting reworked again because of this.

Not going to show the exact camera here, and I think we already deleted the pics taken with it. But as an estimate, for the typical paperback A5-sized page, decent OCR requires the image to be at least around 1MB in resolution (standard aspect ratio).

You May Also Enjoy

less aggressive silence filtering

2 minute read

I modified the aggressiveness from 3 to 1. Training history can be found here for a run which used the LJSpeech mode I trained, and here for a model trained ...

webrtcvad and Shorter Sections

1 minute read

Cheryl and I spent some time taking a bit of a break from training models to try and figure out what it is. One big thing could be an issue with long sentenc...

Blizzard

less than 1 minute read

I decided to try yet another dataset, this time the Blizzard dataset. The tensorboard results can be seen here. With this dataset, it is again successful at ...