sunshaoyan / iseenn Goto Github PK

Since we use urllib to request image from image servers to the central server, sometimes the request fails and some exceptions are not handled.

• Identify the exception
• try something like re-connect after failure

Due to lack of state record of user query actions, the backend needs to perform ranking & re-ranking instead just re-ranking. This contributes to duplicate computations.

Consider recording the result list of initial query.

When the indexed dataset is large, the process of loading index may take too long time. e.g., it costs more than 20 minutes to load 1 million image features.

We should:

1. find out the bottleneck
2. try to load indexes in batches

Because currently we store user uploaded image and database image in different collections, too many codes heavily rely on condition statement as if from_db.

Current difficulty of unifying the two collections:

1. Database images are stored on disk, while user uploaded images are directly in the database
2. We do not want to insert every user uploaded image into the database instantly.

Handle image rotation problem with exif info.

Hi,
I am building Deep Video Analytics which does what you are attempting to do and more.
Thought you would be interested.

https://deepvideoanalytics.com/

Best,
Akshay

Hey,
Your project seems quite interesting to me and I want to understand it.
I need your help for local setup. Could you please provide your e-mail id or any source of contact.
Thanks. :)