Daniel Angelov

@BugaDM , what about smaller backbones like MobilNet? That should have faster inference, at the cost of acc/map.

Agreed, the initial curve is larger. I think the main github README + that docs section should have at least something basic like ``` import tadasets import matplotlib.pyplot as plt...

I want to also point to Mask scoring RCNN https://github.com/zjhuang22/maskscoring_rcnn/blob/master/README.md

Another addition would be - [ ] [Mask R-CNN](https://arxiv.org/abs/1703.06870). code: [https://github.com/CharlesShang/FastMaskRCNN](https://github.com/CharlesShang/FastMaskRCNN) - [ ] FC DenseNet [The One Hundred Layers Tiramisu: Fully Convolutional DenseNets for Semantic Segmentation](https://arxiv.org/abs/1611.09326) code: [https://github.com/SimJeg/FC-DenseNet](https://github.com/SimJeg/FC-DenseNet)

Easiest way is to specify it command line when starting the process via `CUDA_VISIBLE_DEVICES=2,7 python predict.py` GPU ids are 0 based. Otherwise define the model under a block ``` with...

Is there a need for any support on this?

Hi @zhiltsov-max @MihailMihaylov97 This issue may explain my previous experience of Datumaro messing up the categories/annotations forcing me to re-write to a custom converter. Is there any description of this...

At what stage do you call the above mentioned code? Before adding an eventListener that would otherwise call the scan function?

I've done the changes that @Adrian658 and @svepe have suggested and this definitely decreased the amount of connect/disconnect I am observing on the MongoDB container. Setting [CONN_MAX_AGE: None](https://docs.djangoproject.com/en/3.2/ref/databases/#persistent-connections) did not...

I want to add a bit more clarification around this issue as I find it a useful method to support the authentication process. For mongo, one can start it like...