Converting stream.py code for single image

[syedali1621]

Respected sir:
I am trying to convert stream.py code for single image but i am facing many issues can you please provide code for image ?

Thank you

[syedali1621]

Kindly check my code for the single image it is working but the result is not good can you check. Am I missing something in the code or this code is okay?

import tensorflow as tf
tf.logging.set_verbosity(tf.logging.ERROR)
import os
import time
import contextlib
import numpy as np
import cv2
import matplotlib.pyplot as plt

from utils import VideoStream
from tensorflow.keras.models import load_model
from tensorflow.keras.utils import to_categorical
from models import preprocess_input, dice
from config import imshape, model_name, n_classes
from utils import add_masks, crf
with contextlib.redirect_stdout(None):
import pygame

model = load_model(os.path.join('models', model_name+'.model'),
custom_objects={'dice': dice})

img = cv2.imread('075189.png')

plt.imshow(img)

img = cv2.resize(img,(256,256))
img = np.reshape(img,[1,256,256,3])

[0.8, 0.2]

classes = model.predict(img)
print(classes.shape)
#plt.imshow(classes[0, :, :, 0])
plt.imshow(classes[0, :, :, 1])
plt.show()
#cv2.imshow('image',img)
#cv2.waitKey(0)
#cv2.destroyAllWindows()

print (classes)

[seth814]

What do you mean by result is not good? Like it's plotting something, but the output doesn't look correct? Attach an image or something.

[syedali1621]

Thank you for your reply I am attaching result image now results are good but it is not showing RBG result I am segmenting the neck area of T-shirts.

[seth814]

I'm assuming you only have one class then. If I just follow along with the code in stream.py...

img.shape = (256, 256, 3)

tmp = np.expand_dims(img, axis=0)
roi_pred = model.predict(tmp)
roi_mask = roi_pred.squeeze() * 255.0
roi_mask = cv2.cvtColor(roi_mask, cv2.COLOR_GRAY2RGB)
roi_mask = np.array(roi_mask, dtype=np.uint8)
roi_mask = cv2.addWeighted(img, 1.0, roi_mask, 1.0, 0)
plt.imshow(roi_mask)
plt.show()

Let me know if that works. If it does, please post a picture. If it doesn't post the error code.

[syedali1621]

Receiving below error.
module 'numpy' has no attribute 'expand_dim'

[seth814]

Should be np.expand_dims. I edited the original comment.

[syedali1621]

Thank you Now this error.
Invalid number of channels in input image: 'VScn::contains(scn)'where 'scn' is 2

[seth814]

Has something to do with the image channels. I guess the error is something with an invalid shape going into cvtColor, but idk.

[syedali1621]

Somehow i got the RGB result with mask area remove from image but it is showing some noise on image background. Do you have any idea why there is noise? I trained with 256256 images should I train with bigger size images to remove noise?
How to train with bigger size images when I try it gives an error that the image size should be 256256 like.

[seth814]

Training with a bigger image won't do anything.
You can maybe try some of these:
https://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_imgproc/py_morphological_ops/py_morphological_ops.html

Can probably remove the edge at the very top.

[syedali1621]

Thank you yes i am trying them they are improving the results thank you. Also, I need to train on 1024*1024 i changed the values in config.py but receiving below error.

ValueError: Error when checking input: expected input_1 to have shape (1024, 1024, 3) but got array with shape (256, 256, 3)

[seth814]

I probably missed setting a dimension somewhere. Problem is I don't have time to look into this at the moment. Just look at the stack trace or look for where I have 256 in the code and find the conflicts.

[seth814]

Made a notebook for this.

https://github.com/seth814/Semantic-Shapes/blob/master/Single%20Image%20Multi%20Class.ipynb

[syedali1621]

Thank you for the code but i am receiving below error with this code.

File "E:\aithe\projects\virtualfitting\dev\semanticshapes2\utils.py", line 80, in add_masks
val = pred[:,:,i].astype(np.uint8)
IndexError: index 2 is out of bounds for axis 2 with size 2