Feature: Video Inference #200
Conversation
| >>> prediction = model.predict("video.mp4", fps=5, inference_type="object-detection" | ||
| """ | ||
|
|
||
| url = urljoin(API_URL, "/video_upload_signed_url/?api_key=", self.__api_key) |
There was a problem hiding this comment.
I think this API will only return a signed_url for uploading, and user still need to upload the video by himself, right ? @bigbitbus
There was a problem hiding this comment.
Yes, the user can use that returned value to upload a video to that signed url
roboflow/models/video.py
Outdated
|
|
||
| data = response.json() | ||
|
|
||
| if data["success"] != 0 or data["status_info"] != "success": |
There was a problem hiding this comment.
if int(data['status']) == 0:
success
elif int(data['status']) == 1:
still running
else:
failed....will retry it if int(data['retries']) <= 3
|
|
||
| attempts += 1 | ||
|
|
||
| if response != {}: |
There was a problem hiding this comment.
I'd suggest moving this check before sleep
| >>> prediction = model.predict("video.mp4", fps=5, inference_type="object-detection" | ||
| """ | ||
|
|
||
| url = urljoin(API_URL, "/video_upload_signed_url/?api_key=", self.__api_key) |
There was a problem hiding this comment.
Yes, the user can use that returned value to upload a video to that signed url
roboflow/models/video.py
Outdated
| url = urljoin(API_URL, "/video_upload_signed_url/?api_key=", self.__api_key) | ||
|
|
||
| if fps > 5: | ||
| raise Exception("FPS must be less than or equal to 5.") |
There was a problem hiding this comment.
Lets set this to a higher value, up to the frame rate of the video itself.
@PacificDou lets add logic in the video server to use fps = min(video_fps,s infer_fps); and set this actual value into the database object returned to the user when they query the videojob.
There was a problem hiding this comment.
The user needs to know that the FPS can be upto the video frame rate, but given the fps-based pricing, set this value to as low as makes sense to their app. Maybe make it part of the blogpost?
There was a problem hiding this comment.
It would be great if the server raised an exception if FPS is higher. Otherwise, if the user provides a video URL we have to download it in the package to check its FPS before we can do anything.
There was a problem hiding this comment.
The current behavior is that if the video frame-rate < infer_fs, the job fails and the user gets a helpful error message (this check happens when the video finally gets downloaded into the server and processing starts). Unfortunately probing a video to get its true video framerate is not possible at the time the request is made, its part of a batch operation.
Description
This PR adds support for video inference to the Roboflow Python package.
Type of change
How has this change been tested, please provide a testcase or example of how you tested the change?
Will add a description once the draft PR is complete.
Any specific deployment considerations
N/A
Docs
We will publish: