Solved – TensorFlow Object Detection – Multiple objects detection with the pet detector

computer visiondeep learningmachine learningobject detectiontensorflow

I have trained the pet detector from tensorflow object detection with a ssd mobilenet architecture.

The train dataset only contains image with a single objects (1 image = 1 box ).

I would like to detect multiple objects on personnal data. Is it possible to use this model on my data or I need to train a new model with dataset which contains multiple elements ? (for the moment return only 0 or 1 box for 1 image)

More generally, do I need a dataset with images containing multiple objects to have a multiple object detector ?

Thanks

Best Answer

No, you do not need a dataset with many objects on each image in order to train a detector that is able to detect many objects in images. To be more concrete, if you have a dataset with at most one person on each image, it is possible to train a single shot detector that is able to detect several people on an image.

Related Solutions

Object Detection – How to Annotate Negative Samples Using TensorFlow

Short answer:

You don't need negative samples to train your model. Focus on improving your model.

Long answer:

Ideally your detector after being trained on your 2 objects would detect them and place a bounding box around them. When you test it and get wrong results this could be caused by a variety of reasons:

A wrong object is being detected if your detector just misinterpreted as belonging to one of the 2 objects you trained it on.
When an image of your 2 objects is being thrown to the detector but it's not detected, then this also means your detector failed to detect your object.

Both cases could of course happen in the same image and also occur more than once.

As far as I know images with no bounding box in them which would be your "negative examples" cannot be processed by the current model because tfrecord reader just fails.

Negative samples do exist implicitly though. All region of your images that do not correspond to a bounding box is a "negative sample". Defining explicitly "negative samples" by selecting them in a bounding box will create a new class with name 'none'. You will have 3 classes then.

So, to make it simpler focus on your positive examples. If your model fails greatly then something is wrong. Check if:

Your test images differ significantly from the trained ones. If so, try to find images similar to your test images and use them along your previous images to retrain your model.
Examine that you have enough training samples to train your model. Enough is not defined strictly but may mean at least a few hundred images.
Examine that your train model contains enough samples from both objects. If not the misrepresented class will be detected poorly compared to the other one.
Examine that you don't have implicit negative examples of your object that could hinder training process. This means if you detect cars make sure at least all visually qualitative appearances in your image are being labels as cars. If you only label 1 car and also left 2 unlabeled ones then those two cars will hamper your training.

Solved – What are the best methods for reducing false positives in TensorFlow Mask-RCNN object detection framework using transfer learning

A lot of people I see online have been running into the same issue using Tensorflow API. I think there are some inherent problems with the idea/process of using the pretrained models with custom classifier(s) at home. For example people want to use SSD Mobile or Faster RCNN Inception to detect objects like "Person w/ helmet," "pistol," or "tool box," etc. The general process is to feed in images of that object, but most of the time, no matter how many images...200 to 2000, you still end up with false positives when you go actually run it at your desk.

The object classifier works great when you show it the object in its own context, but you end up getting 99% match on every day items like your bedroom window, your desk, your computer monitor, keyboard, etc. People have mentioned the strategy of introducing negative images or soft images. I think the problem has to do with limited context in the images that most people use. The pretrained models were trained with over a dozen classifiers in many variety of environments like in one example could be a Car on the street. The CNN sees the car and then everything in that image that is not a car is a negative image which includes the street, buildings, sky, etc.. In another image, it can see a Bottle and everything in that image which includes desks, tables, windows, etc. I think the problem with training custom classifiers is that it is a negative image problem. Even if you have enough images of the object itself, there isn't enough data of that that same object in different contexts and backgrounds. So in a sense, there is not enough negative images even if conceptually you shouldn't need negative images. When you run the algorithm at home you get false positives all over the place identifying objects around your own room. I think the idea of transfer learning in this way is flawed. We just end up seeing a lot of great tutorials online of people identifying playing cards, Millenium Falcons, etc., but none of those models are deployable in the real world as they all would generate a bunch of false positives when it sees anything outside of its image pool. The best strategy would be to retrain the CNN from scratch with a multiple classifiers and add the desired ones in there as well. I suggest re-introducing a previous dataset from ImageNet or Pascal with 10-20 pre-existing classifiers and add your own ones and retrain it.

Best Answer

Related Solutions

Object Detection – How to Annotate Negative Samples Using TensorFlow

Solved – What are the best methods for reducing false positives in TensorFlow Mask-RCNN object detection framework using transfer learning

Related Question