Italian CLIP
Paper: Contrastive Language-Image Pre-training for the Italian Language
With a few tricks, we have been able to fine-tune a competitive Italian CLIP model with only 1.4 million training samples. Our Italian CLIP model is built upon the Italian BERT model provided by dbmdz and the OpenAI vision transformer.
Do you want to test our model right away? We got you covered! You just need to head to our demo application. The demo also contains all the details of the project, from training tricks to our most impressive results, and much more!
Training data
We considered four main sources of data:
-
WIT is an image-caption dataset collected from Wikipedia (see, Srinivasan et al., 2021).
-
MSCOCO-IT. This image-caption dataset comes from the work by Scaiella et al., 2019.
-
Conceptual Captions. This image-caption dataset comes from the work by Sharma et al., 2018.
-
La Foto del Giorno. This image-caption dataset is collected from Il Post, a prominent Italian online newspaper.
We used better data augmentation, strategic training choices (we have way less data than the original CLIP paper), and backbone-freezing pre-training. For all the details on that, please refer to our demo.
Experiments
Quantitative Evaluation
To better understand how well our clip-italian model works we run an experimental evaluation. Since this is the first clip-based model in Italian, we used the multilingual CLIP model as a comparison baseline.
mCLIP
The multilingual CLIP (henceforth, mCLIP), is a model introduced by Nils Reimers in his sentence-transformer library. mCLIP is based on a multilingual encoder that was created through multilingual knowledge distillation (see Reimers et al., 2020).
Tasks
We selected two different tasks:
- image-retrieval
- zero-shot classification
Reproducibiliy
Both experiments should be very easy to replicate, we share the two colab notebook we used to compute the two results
Image Retrieval
This experiment is run against the MSCOCO-IT validation set (that we haven't used in training). Given in input a caption, we search for the most similar image in the MSCOCO-IT validation set. As evaluation metrics we use the MRR@K.
| MRR | CLIP-Italian | mCLIP |
|---|---|---|
| MRR@1 | 0.3797 | 0.2874 |
| MRR@5 | 0.5039 | 0.3957 |
| MRR@10 | 0.5204 | 0.4129 |
It is true that we used MSCOCO-IT in training, and this might give us an advantage. However the original CLIP model was trained on 400million images (and some of them probably were from MSCOCO).
Zero-shot image classification
This experiment replicates the original one run by OpenAI on zero-shot image classification on ImageNet. To do this, we used DeepL to translate the image labels in ImageNet. We evaluate the models computing the accuracy at different levels.
| Accuracy | CLIP-Italian | mCLIP |
|---|---|---|
| Accuracy@1 | 22.11 | 20.15 |
| Accuracy@5 | 43.69 | 36.57 |
| Accuracy@10 | 52.55 | 42.91 |
| Accuracy@100 | 81.08 | 67.11 |
Our results confirm that CLIP-Italian is very competitive and beats mCLIP on the two different task we have been testing. Note, however, that our results are lower than those shown in the original OpenAI paper (see, Radford et al., 2021). However, considering that our results are in line with those obtained by mCLIP we think that the translated image labels might have had an impact on the final scores.
