Vladimir Zlobin

Hi [baselines/README.md](https://github.com/google/dopamine/blob/master/baselines/README.md) states: > Unsurprisingly, DQN performs much better when trained with 1% of exploratory actions instead of 10% (as used in the original Nature paper). Why is that unsurprising?...

Reuse the link from 22.1 for both models because they have the same weights.

@dkurt, suggested an idea to draw emotion histograms for interactive_face_detection_demo just under bounding boxes to simplify matching between bounding boxes and histograms and decrease the occlusion of image by histograms...

SynthTextDataset augmentation fail

2

Hi. I am trying to run training on synth800k. It fails in `SynthTextDataset` `__transform` method on the line `rectangles = [rectangles[i] for i in selected_poly]` with an error `IndexError: list...

What are metric values for the re-implementation?

1

@xieyufei1993 What results is it possible to achieve with the code on ICDAR 2015/ICDAR 2017? @MaxwellRebo, @Samleo8, @thuyngch, may be you can comment, since you did some updates on your...

The current link is dead

Fix detections scaling

27

1. `LetterBox` uses `round()`: `new_unpad = int(round(shape[1] * r)), int(round(shape[0] * r))`, but `load_image()` had `ceil()`. Align them. `round()` makes more sense to me. 1. Reorder `label['ratio_pad']` to store `(width,...

https://github.com/QwenLM/qwen.cpp/blob/86547f160abf6eb4979a6f246f5177565e662ae4/qwen.cpp#L115 `TextStreamer` waits for the next token before printing punctuation. What's the harm to print it as the last char?