1297rohit / rcnn Goto Github PK

I am not sure, but I guess you are using several different (compared to the paper of RCNN) methods to train and forward the model, maybe you are doing this for making it easy to be understand by us, and I am not sure so I want to point out my guesses to verified it, I will appreciate it if you could reply my guess:
1.use a softmax instead of SVM to make classification
2.I haven't seen bbox regression during training, (maybe I missed...?) and since your code is my first try on R-CNN, I guess the origin model (I mean the one from the paper) use its own bbox regression to get coordinates of box for features-extraction and classification...?
Anyway, your codes help me a lot, Thx!

lenc = MyLabelBinarizer()
Y = lenc.fit_transform(y_new)

TypeErrorTraceback (most recent call last)
in ()
1 lenc = MyLabelBinarizer()
----> 2 Y = lenc.fit_transform(y_new)
3 #lenc = MyLabelBinarizer()
4 #Y = MyLabelBinarizer().fit_transform(y_new)

1 frames
in transform(self, y)
1 class MyLabelBinarizer(LabelBinarizer):
2 def transform(self, y):
----> 3 Y = super().transform(y)
4 if self.y_type_ == 'binary':
5 return np.hstack((Y, 1-Y))

TypeError: super() takes at least 1 argument (0 given)

TypeError: super() takes at least 1 argument (0 given)

Hello..this source code is very helpfull to me because I can study the rcnn concept. I want to ask about the annotation dataset. How can I make the annotation dataset by myself?

Thank you for your help
Regards
Arum

i want to use this for detecting 2 classes: buildings and trees..
can you pls help me with the code.
I am getting problem in preparing the datasets
if i convert my xml file to csv file..the labels corresponding to buildings(1) and trees(0) are lost (labelling tool used is labelImg)
what can be done to mark the labels in the code as it can not be done using the iou values in my case.

please help

When performing this step
trdata = ImageDataGenerator(horizontal_flip=True, vertical_flip=True, rotation_range=90) traindata = X_train.flow(x=X_train, y=y_train) tsdata = ImageDataGenerator(horizontal_flip=True, vertical_flip=True, rotation_range=90) testdata = X_test.flow(x=X_test, y=y_test)

I am getting such an error
MemoryError: Unable to allocate array with shape (26936, 224, 224, 3) and data type float32

How could I go about this?

Hi,
I tried running the notebook but even on rtx-3090(24gb) i wasn't able to run the code due to the high memory requirements of VGG16, so i tried to reduce the image size to (56,56, 3) along VGG16(include_top=Flase, input_shape=(56,56,3), weights='imagenet') and it worked.

it would be great if you can change the transfer learning model to ResNet50 or somewhat similar along with reducing the image size to somewhat smaller in the notebook as not everyone would be able to train this huge model.

Anyways, I appreciate your efforts in explaining the concept on Medium.
Great work !

1. lack of BBox regression?
2. lack of NMS?

I haven't seen it.

Hope for you suggestions!
Thx!

In last cell 'model_loaded' is not defined