chore: migrate examples

[omkar-334]

No description provided.

[aatmanvaidya]

hi, this looks good, we will have to add more examples, I will make a list and drop it here soon, we can see how much of it is possible and how much not

[omkar-334]

hi, this looks good, we will have to add more examples, I will make a list and drop it here soon, we can see how much of it is possible and how much not

Yes, i was looking at the old notebooks, can i work on the legacy operators too? or are they not needed...?

[aatmanvaidya]

@omkar-334 , no need to focus on the legacy operators for example notebooks, here are a few things we can do
I am listing a few notebooks we can write as a part of this task

We should think of all the example jupyter notebooks as independent - this means, the user's don't have to download any data/assests to run the notebook, everything will be taken care in the code of the notebook itself. Even installing the operators should be taken care in the notebook

The goal of each notebook is to demonstrate a real world(ish) use case of feluda

• A t-SNE plot of a set of vidoes (already done)
  • we have already done this using the dimension reduction operator
• Cluster Videos and plot them (already done)
  • we have already done this using the dimension reduction, vid vec rep clip and cluster embeddings operator
• Auto labelling of Videos (already done)
  • we have already done this using the classify zero shot operator
• Cluster Images and plot them
  • the task is to use the image_vec_rep_resnet, dimension reduction and cluster embeddings operators to cluster and visualise a set of images
• Extract Text from Newspaper Images
  • the task is to use the tesseract OCR operator to extract text from newspaper images - you can find some hosted newspaper images here
• Detection of Lewd Images
  • the task is to use the detect_lewd_images operator to show detection capabilities. For the example notebook, you don't have to use sexually explicit images, you can just use normal images.
• Show Perceptual Similarity between Videos using TMK Hash
  • the task is to use the video_hash_tmk to show perceptual similarity between 2 videos by finding their hash values.

---

@omkar-334 do let me know if some of the above things will not be possible in this week, we can scope things according

some hosted media that you can use can be found here - https://github.com/tattle-made/feluda_datasets
if you think more media has to be put here, send them to me, I will push it here, we can use the raw links of these github images in the notebooks.

---

some out of scope advanced examples could be like using a vector DB (like FAISS) to store and search through videos/images

cc @dennyabrain hope the above things regarding example notebooks sound good