yuntai

in `utils.load_weights()` got `valueError: No such layer: batch_normalization_v2` with `2.0.0-beta1` w/o `_v2` it works fine

`pred_bbox = model.predict(image_data)` is much faster; not as fast as your tf1 repo though. > model(x) vs. model.predict(x) When calling model(x) directly, we are executing the graph in eager mode....

this gives a bit of speed-up. very roughly ~ 20 fps to ~ 30 fps on ti1080ti. ``` feature_maps = YOLOv3(input_layer) @tf.function def build(feature_maps): bbox_tensors = [] for i, fm...

for some reason `predict_on_batch(image)` is much faster! (almost twice). tried `predict(image, batch_size=1)` but still slow. with this & `tf.function` above I think, now, the speed is par with that of...

> 请问tf.keras比tf慢多少？现在可以用tf2.0训练自己的数据集吗，如果可以训练 步骤和1.x的一样吗 around 35%

I ran a detector on vdo.avi and dumped out detection result which matches the video clip. https://gist.github.com/yuntai/d0eb58b0eab620db65ac51e326be4c77 using detectron2 (COCO trained faster_rcnn_X_101_32x8d_FPN_3x) from https://github.com/facebookresearch/detectron2/blob/master/MODEL_ZOO.md

> > I ran a detector on vdo.vi and dumped out detection result which matches the video clip. > > https://gist.github.com/yuntai/d0eb58b0eab620db65ac51e326be4c77 > > using detectron2 (COCO trained faster_rcnn_X_101_32x8d_FPN_3x) from >...

> I ran a detector on vdo.avi and dumped out detection result which matches the video clip. > > https://gist.github.com/yuntai/d0eb58b0eab620db65ac51e326be4c77 > > using detectron2 (COCO trained faster_rcnn_X_101_32x8d_FPN_3x) from > https://github.com/facebookresearch/detectron2/blob/master/MODEL_ZOO.md...

https://drive.google.com/file/d/1ADVZyR3BdWUm-saeM6GcFtbw6E2lUcKk/view?usp=sharing

Also, I found this line confusing. `Note that there are two possible variants for each node.` 'Two possible variants' because 0-outputs is not allowed?