I am intending to perform youtube animated video processing ...for detection/classification of activities/actions within the video.. Also I would like to see if the chracters of animated videos resembles real life characters of the same(as in people whose representation is shown in characters)...How do you suggest me to go about with it as I am jus seeing this subject...
How to use a csv file of point cloud with 4 labels? My goal is to train a model to classify point cloud data into the 4 labels, i.e., building, trees, ground and unclassified. Thank you
hey sir? I have one question for you about image resizing, my image dimensions are 1650*3500 height and width but which one is the best size for resizing an image for developing a CNN model?
and alos can u suggest me a way that , I am having 1 folder of food items in that there are 2 folders 1st is fruits and 2nd is snacks and in that both folders I have images of fruits and snacks in respective folders of them but there isnt any management of like apples labels images are in apple folder , banana images are in banana folder , instead of having this arrangement these fruits are all mix in fruit folder and similarly it goes same for snacks folder so also can u suggest me a way how should I deal with this kind of situation with the help of coding
I am doing a cell segmentation and tracking task for one of the CTC differential inference contrast image sequence. I am implementing a U-net architecture. Since the cells could be of different identities (used for cell association in time), it seems to me a multi-class classification problem (each pixel of an image belongs to a certain class/ID). How should I go about defining the last layer of the U-net. Should it be a Conv2D with 'softmax' and 'adam' on 56 filter-depth (where 56 is the total number of cell IDs across the sequence)? Or should I use something like Dense layer?
If you want to segment each cell as unique entity then look into Mask-RCNN. If you are following the process I described then I recommend this to be treated as a binary problem, background and cell. If you define this as multiclass then there must be a way for the system to know the difference between all the cells. But, it fails as all cells may look the same. So in summary, use Mask-RCNN (instance segmentation) or regular binary segmentation. Most tracking algorithms take care of positional information. We will be releasing a tracking workflow on APEER so please stay tuned. If you haven't signed up yet you can do it for free: www.apeer.com
i have an image that needs to be classified into 3 classes ( 0,1,2) images size is ( 224,224,3 ) and total of 180 so for training i am giving 140 ( 140,224,224,3) and labels are changed to categorical so labels size is ( 144 , 3) Inception_model.compile(loss = 'categorical_crossentropy', optimizer = 'Adam', metrics = 'acc') history=Inception_model.fit(x_train,y_train,batch_size=1,epochs=5) ValueError: Shapes (1, 3, 3) and (1, 2048) are incompatible what is this ? how to get out of this ?
TYSM sir for explaining every details in the project .. also does importing test and train is that easy in every dataset like wang or it is differnet for different datasets?
when running feature_extractor = VGG_model.predict(x_train) It give the following error: ValueError: Unexpected result of `predict_function` (Empty batch_outputs). Please use `Model.compile(..., run_eagerly=True)`, or `tf.config.run_functions_eagerly(True)` for more information of where went wrong, or file a issue/bug to `tf.keras`. Anyone can help me out ?
Good video Sreeni!. I have a little something I want to ask: I want to be able to measure the size of cells from histology samples(purple or brown stained). Do you have any pointers on how to go about this business? Would highly appreciate
Here is what I recommend: 1. Normalize images for brightness (if necessary). Please watch my tutorial 122, it may help. 2. Segment nuclei. You can use traditional method if it works (my video 35). If traditional doesn't work then you may need simple machine learning (videos 60 -67). To annotate your images you can use apeer.com/annotate (it is free) 3. After segmentation, extract measurements (video 116)
Sir, I am taking 10 images for testing. So if I want to see all images at a time with printing "this is car" Or this "this is a ship" , then how do I program that.. Actually I trying lot of time to predict the model this way, but I fail. So sir can you help me to how to write the program to get output like that
Not sure what you are asking for. The process in this video explains multiclass classification. If you are asking for custom image classification where you have training images in different folders then watch my video 158.
Impressive explannation. Sir I can't split x_train, x_test, y_train, y_test of my self making image dataset of three classes. How can I load self making dataset and split in x_train, x_test, y_train, y_test keras
Respected sir, i am in need of video that will be showing how to load the data and link csv file for multiclass problem. What if we need to load dataset from pc.. This video is for data that is already in keras dataset. please guide me and give reference of video if already shown. Please reply sir
Sorry, cannot help with your specific issues on laptop as it has too many unknowns but I can suggest you to test it on Google Colab. If it works there then you have to find a way why it isn't working on your specific system. I have checked it on at least 3 different systems, including Raspberry Pi and never faced an issue. If you have any specific message you are getting you can post it here so I (or others) can help.
The goal for this tutorial was to explain multiclass classification and not to find the best model to get the best accuracy for cifar10 data set. The title should have conveyed that message but I will try to do a better job assigning titles to set correct expectations.
