Tensorflow-what better model to perform real's grade training?
I am developing a faculty project to perform the recognition of the real notes through TensorFlow, but the big problem is that the real notes have a lot of similarity between them, especially on the back of the note, another problem is the lack of a dataset with good quality of the new real notes, so it is necessary to create a dataset with good quality images and it may be that at this point I am being flawed.
The configuration of all the TensorFlow in windows and follow the tutorial (https://github.com/EdjeElectronics/TensorFlow-Object-Detection-API-Tutorial-Train-Multiple-Objects-Windows-10and he uses the model of Faster-RCNN-Inception-V2-COCO and one with the image that I'm using in this model, the values of loss are very high to Very High, Even, I would like to know if any of you know of any other model, it would be a good thing for my scenario, and if you have some tips to get around these issues similarity.
2 answers
Assuming you followed the tutorial correctly, the problem may come from two places:
- Data: few data, incomplete data, unbalanced dataset
- hyperparameters
Let's start with data: networks deep learning need a lot of data, but they are a lot even, and that's why they are fashionable, finally we have to do with all our data, you said there are no good real note datasets, so go to a box 2 and 5, Take your mobile phone and start shooting, remember to use the same lighting and background for the different types of note, so the network learns to ignore this information. Or, if you need an easier solution: data augmentation , it consists of techniques to generate new data from the data you already have: invert the image, shift it a few pixels left or right, crop etc...
Hyperparameters: these are the fine adjustments of your Model, learning rate, and other parameters of your optimizer, sometimes tweaking them a little can help the model converge faster
I worked a little with FasteR-CNN and I can say that it is MTO good, it seems that the tutorial is doing transfer learning , retraining only the last layers of the network, but if you are not doing it in real time (processing a video for example) to use a slightly less powerful network
For pattern recognition could use convolutional Neural network (CNN) - based vggnet architecture of some good neural network models, as the CAFFE model recognizes the pattern: https://cocodataset.org/#home abstracting colors and objects like the watermark on the cedulas.