How to load MobileViT for my project

[Sehaba95]

Hi!

I want to use MobileViT and train it using my own dataset (on Google Colab). I couldn't understand how can I use the code in this repository to instantiate just MobileViT model, and use it (I have a very specific data augmentation functions to use).

I am looking for a code to do something like this:

import cvnets.models.classification import MobileViT
import torch

model = MobileViT()
model.load_state_dict(torch.load(PATH))

# Train ...

How can I do that with the original implementation of MobileViT ?

[Amadeus-AI]

Their code of creating model is kind of messy, but I trace their code and found a way to somewhat get rid of it.
First, suggest we have the weight.pt and config.yaml of a model from model_zoo
then we can

import torch
from options.opts import get_training_arguments
from cvnets import get_model

def save_model(*args, **kwargs):
    opts = get_training_arguments()
    model = get_model(opts)
    torch.save(model, 'model_structure.pt')

if __name__ == "__main__":
    save_model()

python .\save_model.py --common.config-file config.yaml

to get the whole specific model structure saved in pt file.

From now on, we can do whatever we want without a lot of unnecessary dependency / information.

import torch

model = torch.load('model_structure.pt')
model.load_state_dict(torch.load('weight.pt'))

# Train ...

Little suggestion to the dev team, some sample code of how to inference your model that list in model_zoo using a single sample picture in 'simple python code' (not some prebuilt executable) will be helpful, like any other modern github model release will do. (https://github.com/openai/CLIP FYI)

[tuobaye11]

how do i use ml-cvnet?
when i use follow command

export CFG_FILE="config/classification/imagenet/resnet.yaml"
cvnets-train --common.config-file $CFG_FILE --common.results-loc classification_results

main_train: Command not found

[acrlife]

how do i use ml-cvnet? when i use follow command

export CFG_FILE="config/classification/imagenet/resnet.yaml" cvnets-train --common.config-file $CFG_FILE --common.results-loc classification_results

main_train: Command not found

I also have this question,are you have a solution?

[LenaDSK]

Their code of creating model is kind of messy, but I trace their code and found a way to somewhat get rid of it. First, suggest we have the weight.pt and config.yaml of a model from model_zoo then we can

import torch
from options.opts import get_training_arguments
from cvnets import get_model

def save_model(*args, **kwargs):
    opts = get_training_arguments()
    model = get_model(opts)
    torch.save(model, 'model_structure.pt')

if __name__ == "__main__":
    save_model()

python .\save_model.py --common.config-file config.yaml

to get the whole specific model structure saved in pt file.

From now on, we can do whatever we want without a lot of unnecessary dependency / information.

import torch

model = torch.load('model_structure.pt')
model.load_state_dict(torch.load('weight.pt'))

# Train ...

Little suggestion to the dev team, some sample code of how to inference your model that list in model_zoo using a single sample picture in 'simple python code' (not some prebuilt executable) will be helpful, like any other modern github model release will do. (https://github.com/openai/CLIP FYI)

[LenaDSK]

I want to fine tune the Mobilevit model for my dataset (only classification model needed). So, I need to download the Mobilevit xxs version, with Imagenet pretrained weights. I have implemented the Read me instructions. I have issues downloading the classification Mobilevit model. Could you please guide me?