added buffalo_l model to deepface

[Raghucharan16]

Tickets

#1187

What has been done

With this PR, buffao_l model will be added to deepface's face recognition models. I ran the face-recognition-how.py for testing the buffalo_l model.

How to test

python tests/face-recognition-how.py

[serengil]

thresholds are missing for various distance metrics for this model here - https://github.com/serengil/deepface/blob/master/deepface/modules/verification.py#L409

[Raghucharan16]

@serengil made the necessary changes could you review once

[serengil]

@serengil made the necessary changes could you review once

@Raghucharan16 Added my review already

[Raghucharan16]

My bad will do that.

[serengil]

Yes, took care of those and updated the code.

I cannot see the changes in the PR - https://github.com/serengil/deepface/pull/1439/files

[Raghucharan16]

Hey @serengil they have provided only google drive links for the model but i doubt this may work or not. what can we do for model's path? push it to somewhere in github?

[serengil]

Hey @serengil they have provided only google drive links for the model but i doubt this may work or not. what can we do for model's path? push it to somewhere in github?

Still you can download it with gdown as https://github.com/serengil/deepface/blob/master/deepface/models/face_detection/Yolo.py#L31

[serengil]

oh ok but how come it got worked on my local machine?

interestingly, it was working fine for me in the first run.

[serengil]

@serengil I got 10/10 linting score and yet it got failed? need to to do anything about bullitins.EXCEPTIONS and BASE_ EXCEPTIONS?

add this

pylint: disable=line-too-long

[serengil]

I found a bug

from deepface import DeepFace

model_name = "Buffalo_L"
ndim = 512

is_verified = DeepFace.verify(
    img1_path = "dataset/img1.jpg",
    img2_path = "dataset/img2.jpg",
    model_name = model_name,
)

img_paths = ["dataset/img1.jpg", "dataset/img2.jpg", "dataset/couple.jpg"]
batch_results = DeepFace.represent(
    img_path=img_paths,
    model_name=model_name,)

assert isinstance(batch_results, list)
assert len(batch_results) == len(img_paths)
for batch_result in batch_results:
    assert isinstance(batch_result, list)
    for result in batch_result:
        assert isinstance(result, dict)
        assert isinstance(result["embedding"], list)
        assert len(result["embedding"]) == ndim

This is because preprocess in Buffalo_L.py is being called with (4, 112, 112, 3) shaped numpy array but the following code block ignores them all.

if len(img.shape) == 4:
    img = img[0]

[serengil]

Failed at linting

************* Module deepface.models.facial_recognition.Buffalo_L
deepface/models/facial_recognition/Buffalo_L.py:67:0: C0301: Line too long (105/100) (line-too-long)
deepface/models/facial_recognition/Buffalo_L.py:95:0: C0304: Final newline missing (missing-final-newline)

please run make lint before pushing something

[Raghucharan16]

I think everything seems fine. let me know if any corrections are needed

[serengil]

LGTM

[Raghucharan16]

@serengil Thanking you for maintaining this cool project. Now i guess you may need to add the new model (buffalo_L) to readme 😅Looking forward to collaborate with you again..

[serengil]

@serengil Thanking you for maintaining this cool project. Now i guess you may need to add the new model (buffalo_L) to readme 😅Looking forward to collaborate with you again..

Will do

Meanwhile, you can create a PR to update readme

[Raghucharan16]

@serengil Thanking you for maintaining this cool project. Now i guess you may need to add the new model (buffalo_L) to readme 😅Looking forward to collaborate with you again..

Will do

Meanwhile, you can create a PR to update readme

Oh sure...will do that..