Strange Depth Output Image

[LulaSan]

Hi everyone, I am not currently using TVM for compatibility issues with colab, so I tried visualize the output without it. This is my code:

import dataloaders.transforms as transforms
from dataloaders.dataloader import MyDataloader
import torch 
from PIL import Image
from PIL import Image
import matplotlib.pyplot as plt
from PIL import Image
import numpy as np
import cv2

img_path = '/content/drive/MyDrive/TESI/FastDepth/fast-depth/cucina.jpg'
imgOrig = np.array(Image.open(img_path))
img = np.asfarray(imgOrig, dtype=np.float32) / 255
img = np.resize(img, (3, 224, 224))
img = np.expand_dims(img, axis=0)
img = torch.from_numpy(img.copy())

print(img.shape) #torch.Size([1, 3, 224, 224])

with torch.no_grad():
  Prediction=model(img)
print(Prediction.shape) #torch.Size([1, 1, 224, 224])`

prediction = cv2.resize(Prediction[0, 0].numpy(),
                       imgOrig.shape[:2][::-1],
                       interpolation=cv2.INTER_CUBIC)
print(f"prediction size is {prediction.shape}")
MAX_DEPTH = prediction.max()
MIN_DEPTH = prediction.min()
prediction = np.abs(prediction)
plt.figure(figsize=(18, 12))
plt.subplot(152)
plt.imshow(prediction, cmap='plasma', vmin=MIN_DEPTH, vmax=MAX_DEPTH)
plt.title('pred depth')
plt.axis('off')
plt.subplot(151)
img_path = '/content/drive/MyDrive/TESI/FastDepth/fast-depth/cucina.jpg'
img = np.array(Image.open(img_path))
plt.imshow(img)
plt.title('orig img')
 `

Is it normal?

P.S. I am using this as trained model

loading model trained

checkpointpath='mobilenet-nnconv5dw-skipadd-pruned.pth.tar'

[LulaSan]

It turned out that this

imgOrig = np.array(Image.open(img_path))
img = np.asfarray(imgOrig, dtype=np.float32) / 255
img = np.resize(img, (3, 224, 224))
img = np.expand_dims(img, axis=0)
img = torch.from_numpy(img.copy())

the resize function, was wrong. I instead did in that way : img= plt.imread(img_path) imgResh = cv2.resize(img, dsize=(224, 224)) img = np.transpose(imgResh, (2,0,1)) img = np.expand_dims(img, axis=0)

[metobom]

@LulaSan Can you share which model did you use?

[LulaSan]

@metobom I used mobilenet-nnconv5dw-skipadd-pruned.pth.tar', loading in this way

import models
import torch
#loading model trained
checkpointpath='mobilenet-nnconv5dw-skipadd-pruned.pth.tar'

if torch.cuda.is_available():
  map_location=lambda storage, loc: storage.cuda()
else:
  map_location='cpu'
checkpoint = torch.load(checkpointpath,map_location=map_location)
if type(checkpoint) is dict:
    start_epoch = checkpoint['epoch']
    best_result = checkpoint['best_result']
    model = checkpoint['model']
    print("=> loaded best model (epoch {})".format(checkpoint['epoch']))