todo

Train:
+ How to update charts (yield)? metric_logger?
+ How to modify checkpoint interval and val interval
+ Visualize predictions (LOGS)
+ metrics, how to add class-wise
+ lr_schedulers? can we plot them? which ones are implemented?
+ augmentations? add imgaug? можно сделать выбор rt-detr / imgaug
+ прочие параметры: ema, ema_decay, etc.
- do not download splits
- lr_schedulers: warmup + scheduler, visualization
- augmentations, handle imgaug in config
- add grad_norm to LOGS and graphs
- Resume training from Team Files checkpoint
- fix re-serve button (config is global and not updated)
- EMA or not EMA
- Stop / pause training
- Consistency in imgsz: Input Size, Resize, 
        cfg.yaml_cfg["HybridEncoder"]["eval_spatial_size"] = size
        cfg.yaml_cfg["RTDETRTransformer"]["eval_spatial_size"] = size


- augmentations in UI, 3 tabs, readonly=False
- charts:
        Loss, Learning Rate, Grad Norm, CUDA Memory;
        AP@IoU=0.50:0.95|maxDets=100
        AR@IoU=0.50:0.95|maxDets=100
        per_class_ap
        per_class_ar



Serve:
- test imgsz support
- use new GUI interface

Others:
- add ONNX support