v0.10.12

Trainer Nodes

Bug-fixes

• Prevent Division-by-Zero Error if all images are already downloaded

v0.10.11

Trainer Nodes

Features

• Add option to set an idle timeout via the environment variables (cf. readme)

Bug-fixes

• Fix bug in image download. This bug has led to new images not being downloaded correctly in some cases

Detector Nodes

Features

• Add API for model version control. A new REST API allows to control the update behavior of the AI model version (cf. readme)

v0.10.10

Detector:

• make outbox use async posts to not block during uploads

v0.10.9

Detector Node:

• Improved outbox behavior. The outbox now uploads images in batches to reduce the workload of the learning loop. If the upload fails, the batch is split until the problematic image can be identified (and deleted)

Trainer Node:

• Fixed wrong handling of SIO re-connection. On connection problems the Trainer node should reset the SIO connection, but there was a bug which is fixed in this version

v0.10.8

What's Changed

• make type hints compatible with Python 3.8
• add actions for testing on Python 3.8

v0.10.7

Detector Node:

• Add endpoint to enable/disable outbox

Trainer Node:

• Change error handling in training stages

v0.10.6

Minor changes:

• We no longer automatically add the Camera-ID as Tag when auto uploading an image during detection.

Fix:

• Change Type hinting to support older python versions

v0.10.5

Fix a rare bug in detection uploading (detections were not uploaded if n_detections == upload_batch_size)

v0.10.4

Bugfix in detect via SocketIO:

• multiple fast detect calls were not throttled and could break the detector.
• this version uses a asyncio lock to make sure the detection processes are not executed in parallel

v0.10.3

Use provided ssl certificate for SIO connection to the learning loop host