inference is not working

[garg-sparsh]

`import torch
from vidar.core.wrapper import Wrapper
from vidar.arch.networks.depth.PackNet import PackNet
from vidar.utils.config import read_config
import numpy as np
import cv2
import torchvision.transforms as transforms
from PIL import Image

def resize_image(image, shape, interpolation=Image.ANTIALIAS):
transform = transforms.Resize(shape, interpolation=interpolation)
return transform(image)

def to_tensor(image, tensor_type='torch.FloatTensor'):
transform = transforms.ToTensor()
return transform(image).type(tensor_type)

cfg = read_config('configs/papers/packnet/inference_packnet.yaml')
net = PackNet(cfg)

rgb = Image.open('/data/vidar/DDAD_images/000081/CAMERA_01/1568648362895305.png')
rgb = resize_image(rgb, (608, 968))
rgb = to_tensor(rgb).unsqueeze(0)
print(rgb)
depth = net(rgb=rgb)['depths']
print(depth)
`

I was able to run the demos/run_network/run_network.py without issues but after applying it to the real image, it throws:
Traceback (most recent call last):
File "demos/run_network/run_network.py", line 48, in
depth = net(rgb=rgb)['depths']
File "/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py", line 1102, in _call_impl
return forward_call(*input, **kwargs)
File "/workspace/vidar/vidar/arch/networks/depth/PackNet.py", line 118, in forward
x4p = self.pack4(x4)
File "/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py", line 1102, in _call_impl
return forward_call(*input, **kwargs)
File "/workspace/vidar/vidar/arch/networks/layers/packnet/packnet.py", line 242, in forward
x = self.pack(x)
File "/workspace/vidar/vidar/arch/networks/layers/packnet/packnet.py", line 150, in packing
x = x.contiguous().view(b, c, out_h, r, out_w, r)
RuntimeError: shape '[1, 256, 38, 2, 60, 2]' is invalid for input of size 2354176

[Diksha-agg]

I did this and I got a list of 4 tensors.

import torch
from PIL import Image
import torchvision.transforms as transforms
import matplotlib.pyplot as plt
import numpy as np

from vidar.arch.networks.depth.MonoDepthResNet import MonoDepthResNet
from vidar.utils.config import read_config

Create network

cfg = read_config('demos/run_network/config.yaml')
net = MonoDepthResNet(cfg)

Load and preprocess the image

image_path = "frame0.jpg" # Replace with the path to your image file
image = Image.open(image_path).convert("RGB")
transform = transforms.Compose([
transforms.Resize((256, 256)), # Resize the image to 256x256 pixels
transforms.ToTensor(), # Convert the image to a PyTorch tensor
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)) # Normalize the image
])
input_tensor = transform(image).unsqueeze(0) # Add a batch dimension to the input tensor
depth = net(rgb=input_tensor)['depths']
print(depth)

Hope this might help!

[Diksha-agg]

You might need to resize your image to the closest power of 2. Like I had 375x1024 size of image. I resized it to 256x1024. It worked.