-
-
Notifications
You must be signed in to change notification settings - Fork 164
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
VisualReplayStrategy
with replay instructions
#610
Conversation
Currently blocked on:
Related: https://community.openai.com/t/gpt4v-and-content-policy-violation/507576/4
Upon retrying:
Related: https://community.openai.com/t/maximum-number-of-images-in-a-gpt-4v-request/480233/4 https://community.openai.com/t/400-errors-on-gpt-vision-api-since-today/534538/23 |
* Cleanup * Add predict for image wrapper. Cleanup * cleanup * Resolve conflicts * Cleanup * Update openadapt/config.py --------- Co-authored-by: Cody DeVilliers <[email protected]> Co-authored-by: Richard Abrich <[email protected]>
VisualReplayStrategy
with replay instructions
This PR implements support for user instructions via
python -m openadapt.replay <ReplayStrategy> --instructions <natural language replay instructions>
For each processed event, if it is a mouse event:
- segment the event's active window
- get a natural language description of each element
Get a natural language description of what the active element should be given the
replay instructions.
For each modified event:
a. Convert descriptions to coordinates
If it is a click, scroll, or the last in a sequence of isolated events:
b. Replay modified event
See prompts for details:
6x8.mov
Raw image:
Segmented image:
Raw masks:
Refined masks:
Masked image descriptions:
gpt-4-vision-preview
:gpt-4-turbo-2024-04-09
:gemini-1.5-pro-latest
:claude-3-opus-20240229
:Event location: