Dynamic View Synthesis from Dynamic Monocular Video

Trying to convert the model to onnx model, but got error

File "to_onnx.py", line 72, in export_model(model, img_input, export_model_name) File "to_onnx.py", line 30, in export_model torch.onnx.export(model, input, export_model_name, verbose=False, export_params=True, opset_version=11) File "C:\Users\yyyy\Anaconda3\envs\torchreid\lib\site-packages\torch\onnx_init_.py", line 148, in export strip_doc_string, dynamic_axes, keep_initializers_as_inputs) File "C:\Users\yyyy\Anaconda3\envs\torchreid\lib\site-packages\torch\onnx\utils.py", line 66, in export dynamic_axes=dynamic_axes, keep_initializers_as_inputs=keep_initializers_as_inputs) File "C:\Users\yyyy\Anaconda3\envs\torchreid\lib\site-packages\torch\onnx\utils.py", line 416, in _export fixed_batch_size=fixed_batch_size) File "C:\Users\yyyy\Anaconda3\envs\torchreid\lib\site-packages\torch\onnx\utils.py", line 279, in _model_to_graph graph, torch_out = _trace_and_get_graph_from_model(model, args, training) File "C:\Users\yyyy\Anaconda3\envs\torchreid\lib\site-packages\torch\onnx\utils.py", line 236, in _trace_and_get_graph_from_model trace_graph, torch_out, inputs_states = torch.jit._get_trace_graph(model, args, _force_outplace=True, return_inputs_states=True) File "C:\Users\yyyy\Anaconda3\envs\torchreid\lib\site-packages\torch\jit_init.py", line 277, in _get_trace_graph outs = ONNXTracedModule(f, _force_outplace, return_inputs, return_inputs_states)(*args, **kwargs) File "C:\Users\yyyy\Anaconda3\envs\torchreid\lib\site-packages\torch\nn\modules\module.py", line 532, in call result = self.forward(*input, **kwargs) File "C:\Users\yyyy\Anaconda3\envs\torchreid\lib\site-packages\torch\jit_init.py", line 332, in forward in_vars, in_desc = _flatten(args) RuntimeError: Only tuples, lists and Variables supported as JIT inputs/outputs. Dictionaries and strings are also accepted but their usage is not recommended. But got unsupported type numpy.ndarray

to_onnx.py

import os
import glob
import torch
import utils
import cv2

from torchvision.transforms import Compose
from models.midas_net import MidasNet
from models.transforms import Resize, NormalizeImage, PrepareForNet

import onnx
import onnxruntime

def test_model_accuracy(export_model_name, raw_output, input):    
    ort_session = onnxruntime.InferenceSession(export_model_name)

    def to_numpy(tensor):
        return tensor.detach().cpu().numpy() if tensor.requires_grad else tensor.cpu().numpy()

    # compute ONNX Runtime output prediction
    ort_inputs = {ort_session.get_inputs()[0].name: to_numpy(input)}
    ort_outs = ort_session.run(None, ort_inputs)	

    # compare ONNX Runtime and PyTorch results
    np.testing.assert_allclose(to_numpy(raw_output), ort_outs[0], rtol=1e-03, atol=1e-05)

    print("Exported model has been tested with ONNXRuntime, and the result looks good!")		

def export_model(model, input, export_model_name):
    torch.onnx.export(model, input, export_model_name, verbose=False, export_params=True, opset_version=11)	
    onnx_model = onnx.load(export_model_name)    
    onnx.checker.check_model(onnx_model)
    graph_output = onnx.helper.printable_graph(onnx_model.graph)
    with open("graph_output.txt", mode="w") as fout:
        fout.write(graph_output)
		
device = torch.device("cpu")

 # load network
model_path = "model.pt"
model = MidasNet(model_path, non_negative=True)

transform = Compose(
        [
            Resize(
                384,
                384,
                resize_target=None,
                keep_aspect_ratio=True,
                ensure_multiple_of=32,
                resize_method="lower_bound",
                image_interpolation_method=cv2.INTER_CUBIC,
            ),
            NormalizeImage(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
            PrepareForNet(),
        ]
)

model.to(device)
model.eval()

img = utils.read_image("input/line_up_00.jpg")
img_input = transform({"image": img})["image"]

# compute
#with torch.no_grad():
sample = torch.from_numpy(img_input).to(device).unsqueeze(0)
print("sample type = ", type(sample), ", shape of sample = ", sample.shape)
print(sample)	
prediction = model.forward(sample)
export_model_name = "midas.onnx"	
export_model(model, img_input, export_model_name)

Environment:

pytorch 1.4.0(installed by anaconda) os is windows 10 64bits

Hej, I dont think this is a issue, sorry for posting like this. But the image that goes through you model is really slow. Do you have a method for speeding it up? Sorry again for posting it as a issue, but dont know how else to make contact.

Hello! Thanks for you work!

I have two questions:

1. What is the .pmf format and what is it used for?
2. While opening .png depth maps how to convert them into float32 in meters units?

Your install.sh script causes every single Python project using your module to break. It is used by some Automatic1111 extensions and NONE of them work unless manually copying the externals/Next_ViT into the Web UI root folder.

Error loading script: depthmap_for_depth2img.py Traceback (most recent call last): File "X:\StableDiffusion\webui\modules\scripts.py", line 195, in load_scripts module = script_loading.load_module(scriptfile.path) File "X:\StableDiffusion\webui\modules\script_loading.py", line 13, in load_module exec(compiled, module.dict) File "X:\StableDiffusion\webui\extensions\depthmap2mask\scripts\depthmap_for_depth2img.py", line 11, in from repositories.midas.midas.dpt_depth import DPTDepthModel File "X:\StableDiffusion\webui\repositories\midas\midas\dpt_depth.py", line 5, in from .blocks import ( File "X:\StableDiffusion\webui\repositories\midas\midas\blocks.py", line 21, in from .backbones.next_vit import ( File "X:\StableDiffusion\webui\repositories\midas\midas\backbones\next_vit.py", line 8, in file = open("./externals/Next_ViT/classification/nextvit.py", "r")

When an app uses your module, it will expect the externals/Next_ViT directory at the app root. You cannot import a module in Python to a fixed location because that will break the dependency, since you're installing it in the wrong location. Your module won't find the module, because the main app installes it to it's own root, not to your module's root.

Anyone using your module will have to manually copy the externals folder to their own app's root folder, which is really unacceptable.

You should clone it to your own project's root using python git and load it by instead of using

file = open("./externals/Next_ViT/classification/nextvit.py", "r")

in midas/backbones/next_vit.py Use

file = open("../../externals/Next_ViT/classification/nextvit.py", "r")

Also, a shell script to run a git command? Really?? You think nobody uses another operating system? This is why I say you MUST use Python's own git module.

Hi, Before launching your code, I would like to understand how to get a maximum depth by numeric, and the x/y axis. Could I have your advice on it? If you could tell me the specific key function where generates the maximum depth, it would be really helpful.

Allow me to add one more question, is your code available to use on the google colaboratory?

I'm looking forward to hearing from you soon!

Hi! I have runtime problems:

...
File "/home/etienne/.cache/torch/hub/facebookresearch_WSL-Images_master/hubconf.py", line 23, in _resnext
    model = ResNet(block, layers, **kwargs)
TypeError: __init__() got an unexpected keyword argument 'groups'

Searching on this, it seems it's because of a bad torchvision version. I do not succeed in solving all version conflicts. Is it possible to have a minimal set of command lines to install all in the right version?

A "prediction" gives the following:

[[2496.0127 2495.973 2495.9888 ... 855.7698 855.57666 856.0468 ] [2495.9575 2495.9158 2495.9329 ... 855.4036 855.20917 855.68256] [2495.9797 2495.9387 2495.9556 ... 855.55426 855.36035 855.83234] ... [3245.7551 3245.7756 3245.7664 ... 2852.5774 2852.4922 2852.702 ] [3245.7275 3245.7478 3245.739 ... 2852.4827 2852.397 2852.6072 ] [3245.7974 3245.8179 3245.809 ... 2852.7156 2852.6309 2852.8398 ]]

What are the units of these numbers m, mm, ft? Of course those numbers aren't disparities (since the images aren't that wide). So what do these numbers represent? How to convert this prediction to actual depth given camera intrinsics? Thanks

Hello

I got this error. Did I put the .pt file in wrong location? (see attachment)

initialize device: cuda Loading weights: model-f46da743.pt Using cache found in C:\Users\gregb/.cache\torch\hub\facebookresearch_WSL-Images_master Traceback (most recent call last): File "run.py", line 105, in run(INPUT_PATH, OUTPUT_PATH, MODEL_PATH) File "run.py", line 29, in run model = MidasNet(model_path, non_negative=True) File "C:\Users\gregb\Documents\Python\MiDaS\midas\midas_net.py", line 47, in init self.load(path) File "C:\Users\gregb\Documents\Python\MiDaS\midas\base_model.py", line 11, in load parameters = torch.load(path) File "C:\Users\gregb\anaconda3\envs\3DP\lib\site-packages\torch\serialization.py", line 525, in load with _open_file_like(f, 'rb') as opened_file: File "C:\Users\gregb\anaconda3\envs\3DP\lib\site-packages\torch\serialization.py", line 212, in _open_file_like return _open_file(name_or_buffer, mode) File "C:\Users\gregb\anaconda3\envs\3DP\lib\site-packages\torch\serialization.py", line 193, in init super(_open_file, self).init(open(name, mode)) FileNotFoundError: [Errno 2] No such file or directory: 'model-f46da743.pt'

@ranftlr Thank you for the work. I'm trying to apply it with Myriad X VPU. So I would like to ask whether the unknown scale and shift mentioned in #36 are linear parameters? For example, in each frame, I can find a linear equation like "P = D * scale + shift" to project the values of depth maps "D" to the physical absolute measurements "P" according to putting a known scale ruler in the view, right ?

Hello! I am using the following jupiter notebook https://github.com/deforum/stable-diffusion/blob/main/Deforum_Stable_Diffusion.ipynb

When setting up the environment, the following line causes problems https://github.com/isl-org/MiDaS/blob/eaa249ff68fde82cad78982d0a36523e3ebfbfdf/midas/backbones/next_vit.py#L8

It looks like it wasn't intended to be used this way.

Hi, is there any evaluation code for MiDaS?

MiDaS can predict a robust inverse-depth of a single image, but how can I eval on datasets with ground truth depth like KITTI? Should I convert predict disparity to depth and evaluate in the depth space, or convert ground truth depth to disparity and evaluate in the disparity space?

I downloaded the official validation set of KITTI, and convert gt_depth to gt_disparity with:

gt_disparity = 1 / (gt_depth + 1e-8)
gt_disparity[gt_depth==0]=0

after performing alignment in disparity space and evaluating on midas_v3.0_dpt-large on KITTI, I got the performance:

KITTI AbsRel : 10.0
KITTI delta > 1.25 : 10.1

It seems that my evaluation code is not so accurate. Could you please provide your evaluation code for MiDas? Thank you so much.

Hi all,

We're the maintainers of torch.hub and it looks like executing the code in the MiDaS example fails with the following error:

QObject::moveToThread: Current thread (0x5f956b80) is not the object's thread (0x601f7590).
E           Cannot move to target thread (0x5f956b80)
E           
E           qt.qpa.plugin: Could not load the Qt platform plugin "xcb" in "/home/circleci/miniconda3/lib/python3.8/site-packages/cv2/qt/plugins" even though it was found.
E           This application failed to start because no Qt platform plugin could be initialized. Reinstalling the application may fix this problem.
E           
E           Available platform plugins are: xcb, eglfs, minimal, minimalegl, offscreen, vnc, webgl.

You can find more details about the execution here

Would you mind submitting a fix either in your own repo if relevant, or directly to the MiDaS torchhub file? Thank you!

(I suspect that the problem may be that Qt is required to run the model somehow, while our CI job is headless so the X server isn't available - I could be completely wrong though).

Hi，when building the Docker image with :docker build -t midas . the following error occured: GPG error: https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64 InRelease: The following signatures couldn't be verified because the public key is not available: NO_PUBKEY A4B469963BF863CC #6 3.925 E: The repository 'https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64 InRelease' is not signed.

try fix with: sudo apt-key adv --keyserver keyserver.ubuntu.com --recv-keys A4B469963BF863CC but failed.

Do you have any idea?

So basically what I’ve tried doing is converting a 2D video into a 3D video with anaglyph.

1. I render a video clip into a PNG sequence using Adobe Media Encoder
2. I run the PNG sequence using MiDas
3. I insert the both DepthMap and the normal sequence in after effects as PNG sequence
4. I Displace Map the depth and Right shift +5 and left shift -5 on another copy, creating 2 different perspectives
5. I color the right one Cyan and the left one Magenta and apply Multiply blending mode
6. Color correcting using adjustment layer
7. Then render to final result

the issue is that the Depth sequence rendered with MiDas is uneven, it’s flickering and it’s slight light changes, although the original clip has a flat exposure throughout the video.

I’m using the Large MiDas file to render. Was wondering if there is another more ”accurate” way to do this? It seems to me the results other people are getting is way more on-point whilst mines is kind of blurry and uneven throughout the sequence!

Hello, In my dedicated python environment, I tried with latest timm and forced 0.4.5 without luck:

pip install timm==0.4.5
Defaulting to user installation because normal site-packages is not writeable
Collecting timm==0.4.5
  Downloading timm-0.4.5-py3-none-any.whl (287 kB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 287.4/287.4 kB 2.8 MB/s eta 0:00:00
Requirement already satisfied: torch>=1.4 in /home/pm/.local/lib/python3.10/site-packages (from timm==0.4.5) (1.12.1)
Requirement already satisfied: torchvision in /home/pm/.local/lib/python3.10/site-packages (from timm==0.4.5) (0.13.1)
Requirement already satisfied: typing-extensions in /home/pm/.local/lib/python3.10/site-packages (from torch>=1.4->timm==0.4.5) (4.3.0)
Requirement already satisfied: pillow!=8.3.*,>=5.3.0 in /usr/lib/python3/dist-packages (from torchvision->timm==0.4.5) (9.0.1)
Requirement already satisfied: numpy in /usr/lib/python3/dist-packages (from torchvision->timm==0.4.5) (1.21.5)
Requirement already satisfied: requests in /usr/lib/python3/dist-packages (from torchvision->timm==0.4.5) (2.25.1)
Installing collected packages: timm
Successfully installed timm-0.4.5
(midas) pm@pop-os:~/Documents/github$ python run.py --model_type dpt_large
Traceback (most recent call last):
  File "run.py", line 11, in <module>
    from midas.dpt_depth import DPTDepthModel
  File "/home/pm/Documents/github/midas/dpt_depth.py", line 6, in <module>
    from .blocks import (
  File "/home/pm/Documents/github/midas/blocks.py", line 4, in <module>
    from .vit import (
  File "/home/pm/Documents/github/midas/vit.py", line 3, in <module>
    import timm
ModuleNotFoundError: No module named 'timm'

Any idea?

I'm having trouble figuring out how to define a location to store downloaded models, and load them. Every run it downloads models, and the probability of connections ended by peers, and other stuff is high. I'd like to prevent this, and keep bandwidth consumption load. I couldn't find anything in the documentation regarding model paths.

I've tried to run MiDaS on Apple silicon with pytorch nightly with mps (metal performance shader) support.

I've change two lines of code to get this working: 28 - device = torch.device("cuda" if torch.cuda.is_available() else "mps") and 127 - mode="bilinear",

The change to "bilinear" is necessary because bicubic is not supported currently for the M1 native implementation. Influence on output quality is neglectable in my experience (I did a few comparing tests on "CPU" with only this parameter changed)

This way I could speed up inference by a whopping factor of 6. But: Output quality detoriates to not-usable. I've got virtually no experience with python and pytorch - I'm basically a hobbyist creator and dad who wants to have a depth map on selected gopro footage of his son to fake depth of field into it. So i do not know where to look at.

Legacy Midas V2 seems to work okay and delivers the same output on CPU and MPS: CPU MPS

DPT however, output becomes unusable with mps: MPS

CPU

Any ideas? Can someone give me a hint?