Predict on new images

[billsun9]

Hello,

Given the provided model (model_final.pth), how would I predict on new images?

[kushalchordiya216]

I would also like to know the answer to this

[xuecho]

I also have the question, could anyone answer please?

[hellollw]

I am looking forward to using this model to train my own dataset and predict the unseen categories without fine-tuning!

[billsun9]

Dear all,

I have figured out how to predict on new images by reading the Detectron2 documentation. Essentially, you need to run demo.py from detectron2 https://github.com/facebookresearch/detectron2 after altering the configuration files to account for FewX.

@hellollw I believe you can write a training script by following detectron2 as well

Here is an example prediction from the model provided:

[hellollw]

Dear all,

I have figured out how to predict on new images by reading the Detectron2 documentation. Essentially, you need to run demo.py from detectron2 https://github.com/facebookresearch/detectron2 after altering the configuration files to account for FewX.

@hellollw I believe you can write a training script by following detectron2 as well

Here is an example prediction from the model provided:

Oh, your answer is very helpful to me! So I just need to first train the FewX model using the given code, and then run the demo.py after altering the model path? But if I want to train the model with my own dataset, should I alter FewX code or just use the detectron2 code?

[Mas-Y]

Dear all,

I have figured out how to predict on new images by reading the Detectron2 documentation. Essentially, you need to run demo.py from detectron2 https://github.com/facebookresearch/detectron2 after altering the configuration files to account for FewX.

@hellollw I believe you can write a training script by following detectron2 as well

Here is an example prediction from the model provided:

I tried running inference using the model_final.pth provided in this repo the inference runs fine. but when i try with the intermediate saved model saved during my training on coco2017, i am getting blank predictions. The .pth i tried are: model_0029999.pth, model_0059999.pth. Could you please help what could be the issue?

[selimlouis]

@billsun9 how exactly did you alter the config to train with demo.py?

Also did you manage to actually test the few-shot capability on unseen categories? I do not understand how exactly to give the model support images and then query images to detect the new objects

Help would be really appreciated

[xiaofeng-c]

i try to use the model test the image given,but how we write a ".yml" file to inference new images?

[pcyyo]

Dear all,

I have figured out how to predict on new images by reading the Detectron2 documentation. Essentially, you need to run demo.py from detectron2 https://github.com/facebookresearch/detectron2 after altering the configuration files to account for FewX.

@hellollw I believe you can write a training script by following detectron2 as well

Here is an example prediction from the model provided:

hello ！can you tell me how to modify the config file ? thank you

[Lei-1998]

Dear all,

I have figured out how to predict on new images by reading the Detectron2 documentation. Essentially, you need to run demo.py from detectron2 https://github.com/facebookresearch/detectron2 after altering the configuration files to account for FewX.

@hellollw I believe you can write a training script by following detectron2 as well

Here is an example prediction from the model provided:

Please tell me how you debug it? If you can, please send me the code. My email is [email protected]