(feat) Face Detector and OCR Extractor UDFs #303
Conversation
|
Do we need |
|
No, we don't need them. In fact, we are not yet enforcing the |
|
I added defaults for |
| """ | ||
|
|
||
| copy = torch.squeeze(frames) | ||
| transform = T.ToPILImage() |
There was a problem hiding this comment.
This converison is wierd. Are we doing it frame by frame?
There was a problem hiding this comment.
Should we do it in some other manner?
pil_image = T.ToPILImage(torch.squeeze(frames))
bboxes, scores = self.model.detect(img=pil_image)
There was a problem hiding this comment.
model takes in ndarray or PIL image
There was a problem hiding this comment.
The concern is that we do a few transformations https://github.com/georgia-tech-db/eva/blob/e6beba4d5da7902108aed3c7fa6b0f366cfd0219/eva/udfs/pytorch_abstract_udf.py#L47
before feeding it to the UDF. (ndarray->PIL->Tensor)
Now inside the UDF, we also repeat those transformations (tensor -> PIL).
There was a problem hiding this comment.
We need to provide a cleaner interface.
jarulraj
changed the title
-- Added UDF and an integration test case