add SAM model #1847
add SAM model #1847
Conversation
|
@divyashreepathihalli I think you need to move this to the keras-hub directory now after the rename |
done |
mattdangerw
left a comment
mattdangerw
left a comment
There was a problem hiding this comment.
Thanks! Left some comments, look like they might show up in the old location sadly.
| self.bottom_right_corner_embed = keras.layers.Embedding( | ||
| 1, hidden_size, name="bottom_right_corner_embed" | ||
| ) | ||
| self.not_a_point_embed = keras.layers.Embedding( |
There was a problem hiding this comment.
what is a "not a point embed?
There was a problem hiding this comment.
Here is the original implementation of this. - self.not_a_point_embed = nn.Embedding(1, embed_dim)
When you use SAM without providing explicit point prompts, , the model still needs some input to represent the "absence" of point information. This is where not_a_point_embed is used. not_a_point_embed acts like a placeholder or default value when you don't specify point prompts. It allows the model to function correctly even without explicit point guidance.
SamanehSaadat
left a comment
SamanehSaadat
left a comment
There was a problem hiding this comment.
Thanks, Divya!
Left some nit comments!
| (batch_size, 0, image_size, image_size, 1) | ||
| ), | ||
| } | ||
| # todo: update preset name |
There was a problem hiding this comment.
Is this for later? Or the presets are uploaded and this can be updated?
There was a problem hiding this comment.
This is for later. The weights are yet to be added. I will update the name once we have them uploaded.
| query=queries, value=queries, key=queries | ||
| ) | ||
| else: | ||
| queries_with_pe = queries + query_pe |
There was a problem hiding this comment.
What does pe stand for here?
There was a problem hiding this comment.
positional embeddings. I updated the names wherever we have _pe
3 participants
No description provided.