l am following this tutorial for Multiclass semantic segmentation and i face a problem. the lable in 8:26 code page my "label in the mask " = [0 38 75]. is it okay to play this segmentation code? My n_classes=3 (background, man, monkey) I'm not good at english sorry If my "label in the mask " is wrong, Can you tell me what video i watch to fix the problem thank you for your help
sir i am always follow your code, my question is, if i custom the mode using VGG16 with weights="imagenet" i am alwas get error with my shape image, should i use rgb image with number chanel 3 or 1 or 0 for my dataset image and my dataset masking?
Thanks Sini for such wonderful contents. I have one doubt, i am working on a healtcate dataset on a problem of instance segmentation to detect 3 different types of neurons. the masks that I have has background label 0 then 3 classes of neuronal cell types 1, 2 , 3. So I have to consider 4 classes right? I actually tried considering background also as class and trained model and got good IoU metric around 0.83, but the mean IoU is very small, less than 0.1. I am not sure what wrong I am doing here. Can you suggest how i can go ahead to this problem?
Dear, Sir I am continuously following your tutorials. You have a solution to my every problem. I will be using your methods on meteorite BSE images. And try to segment carbonaceous matter from those images just like you segment sandstones, clay and pores.
Good to know Rahul. One challenge with meteorite BSE images is that you may not have well defined borders for certain minerals. For example, you may find a gradual grey level change due to zoning in Olivines. Good luck.
18:38 My dataset is some pictures of cars in 13 model classes, best model i could build was loss: 0.030 - accuracy: 0.988 - val_loss: 1.317 - val_accuracy: 0.806 & loss: 0.100 - accuracy: 0.966 - val_loss: 1.077 - val_accuracy: 0.768 but my mean IoU is near 0, and only class background and class 1 has IoU > 0 like wtf others all exactly 0 I generated 50 new random images for each image also, IDK why my model gets overfit. My bridge has only 128 filters and encode1,2,3,4 each has 8,16,32,64 filters
When I use the provided stacked image (128_patches) I get an error we I run the following line: image_dataset = np.expand_dims(image_dataset, axis = 3). The error is : AxisError: axis 3 is out of bounds for array of dimension 2. I could solve this by patching the images myself. Yet the validation Accuracy and Validation loss were so bad. I appreciate any help.
![BLACK BAG - Official Trailer [HD] - Only in Theaters March 14](http://i.ytimg.com/vi/Du0Xp8WX_7I/mqdefault.jpg)
I am eagerly waiting to see a complete tutorial (Theory + code ) on Transformer-based UNet like TransUNet, Great job Dr.
Haven't watch the whole video yet but I've already learned so much, thank you for the great content
I am very excited to start learning about 3D semantic segmentation, thank you Sreeni :)
You are welcome 😊
You saved my life, thank you so much....
It's an awesome implementation. Good job.
Thank you.
In the link provided for dataset , I found tifstack but there are no individual tif images in the specified folder
Thanks for your wonderful videos, I got a question, I am trying to read the dataset but is as a stack tiff format so how to read it?
Thank you very much.
l am following this tutorial for Multiclass semantic segmentation and i face a problem. the lable in 8:26 code page my "label in the mask " = [0 38 75]. is it okay to play this segmentation code?
My n_classes=3 (background, man, monkey)
I'm not good at english sorry
If my "label in the mask " is wrong, Can you tell me what video i watch to fix the problem
thank you for your help
THANKS prof.
in my code it gives me this error: axis 3 is out of bounds for array of dimension 3
Hi sir,
how can I get to know which class is predicted?
sir how do we plot confusion matrix
sir my mask show
Labels in the mask are : [ 0 10 154 254]
so how i can covert into classes 1,2,3,4
Fantastic video!
If I have class 0,1,2,3,4 and 0 as background class..how to delete that 0 class while doing label encoder
You need a label for every pixel that goes through the network. This means, you should not delete the pixels from background with a value of 0.
Can you make a video for semantic segmentation by training a data using U-net ?
Instead of 200 images if I want to load 5000 images but the memory is not allowing that. I have 24 GB of RAM still its not possible
sir i am always follow your code, my question is, if i custom the mode using VGG16 with weights="imagenet" i am alwas get error with my shape image, should i use rgb image with number chanel 3 or 1 or 0 for my dataset image and my dataset masking?
really useful tutorial videos you have made, thanks sir. I am not having a mask folder for my dataset. without mask folder how to do the segmentation?
Sri please make a video on instance segmentation.
Thanks Sini for such wonderful contents. I have one doubt, i am working on a healtcate dataset on a problem of instance segmentation to detect 3 different types of neurons. the masks that I have has background label 0 then 3 classes of neuronal cell types 1, 2 , 3. So I have to consider 4 classes right? I actually tried considering background also as class and trained model and got good IoU metric around 0.83, but the mean IoU is very small, less than 0.1. I am not sure what wrong I am doing here. Can you suggest how i can go ahead to this problem?
Dear, Sir I am continuously following your tutorials. You have a solution to my every problem. I will be using your methods on meteorite BSE images. And try to segment carbonaceous matter from those images just like you segment sandstones, clay and pores.
Good to know Rahul. One challenge with meteorite BSE images is that you may not have well defined borders for certain minerals. For example, you may find a gradual grey level change due to zoning in Olivines. Good luck.
how to download dataset?
any one can send link
drive.google.com/file/d/1HWtBaSa-LTyAMgf2uaz1T9o1sTWDBajU/view
18:38 My dataset is some pictures of cars in 13 model classes,
best model i could build was loss: 0.030 - accuracy: 0.988 - val_loss: 1.317 - val_accuracy: 0.806 & loss: 0.100 - accuracy: 0.966 - val_loss: 1.077 - val_accuracy: 0.768
but my mean IoU is near 0, and only class background and class 1 has IoU > 0 like wtf others all exactly 0
I generated 50 new random images for each image also, IDK why my model gets overfit. My bridge has only 128 filters and encode1,2,3,4 each has 8,16,32,64 filters
shouldn't 17:27 be IOU_keras.update_state(np.argmax(y_test, axis=3), y_pred_argmax) ??
I still get bad results but it's less wierd
I just forgot to normalize my test data before doing prediction and it took me forever to realize it
When I use the provided stacked image (128_patches) I get an error we I run the following line: image_dataset = np.expand_dims(image_dataset, axis = 3). The error is : AxisError: axis 3 is out of bounds for array of dimension 2.
I could solve this by patching the images myself. Yet the validation Accuracy and Validation loss were so bad.
I appreciate any help.
l have same problem. did you solve and how to get data?
I ran into the following error
ResourceExhaustedError Traceback (most recent call last)
in ()
----> 1 history = model.fit(X_train, y_train_cat,
2 batch_size = 16,
3 verbose=1,
4 epochs=5,
5 validation_data=(X_test, y_test_cat),
1 frames
/usr/local/lib/python3.10/dist-packages/tensorflow/python/eager/execute.py in quick_execute(op_name, num_outputs, inputs, attrs, ctx, name)
50 try:
51 ctx.ensure_initialized()
---> 52 tensors = pywrap_tfe.TFE_Py_Execute(ctx._handle, device_name, op_name,
53 inputs, attrs, num_outputs)
54 except core._NotOkStatusException as e:
ResourceExhaustedError: Graph execution error: