vinid commited on
Commit
ec881bb
1 Parent(s): 624b826

update text

Browse files
Files changed (2) hide show
  1. introduction.md +12 -0
  2. static/img/improvements.png +0 -0
introduction.md CHANGED
@@ -119,6 +119,18 @@ didn't go well. Eventually, the thing that worked out the best was fixing the lo
119
  is used after the computation of the similarity between the images and the texts in CLIP (see the code [here](https://github.com/clip-italian/clip-italian/blob/master/hybrid_clip/modeling_hybrid_clip.py#L64)).
120
  We got this idea from Nils' [video](https://youtu.be/RHXZKUr8qOY) on sentence embeddings.
121
 
 
 
 
 
 
 
 
 
 
 
 
 
122
  # Scientific Validity
123
 
124
  ## Quantitative Evaluation
 
119
  is used after the computation of the similarity between the images and the texts in CLIP (see the code [here](https://github.com/clip-italian/clip-italian/blob/master/hybrid_clip/modeling_hybrid_clip.py#L64)).
120
  We got this idea from Nils' [video](https://youtu.be/RHXZKUr8qOY) on sentence embeddings.
121
 
122
+ ### Effect
123
+
124
+ The following picture showcase the effect that this edits have had on our loss:
125
+
126
+ <img src="https://huggingface.co/spaces/clip-italian/clip-italian-demo/raw/main/static/img/improvements.png" alt="drawing" width="600"/>
127
+
128
+ The purple line is the original training, you can see how many steps we needed to get the loss down. Yellow line is the
129
+ loss with the new optimizer, it is **striking** to see the time we save from this addition! Blue line shows the results when
130
+ fixed scaling is added with the new optimization. Finally, we added the backbone freezing part and you can see the
131
+ results in the light blue loss.
132
+
133
+
134
  # Scientific Validity
135
 
136
  ## Quantitative Evaluation
static/img/improvements.png ADDED