chamecall

I've made the appropriate normalization. Check [the colab](https://colab.research.google.com/drive/1uNcohQhdU2fNNNeQ-xxtM1jxta099_3i?usp=sharing). [TCGA_CS_4944_20010208.zip](https://github.com/mateuszbuda/brain-segmentation-pytorch/files/7010034/TCGA_CS_4944_20010208.zip)

have you found any solution for the case?

> Hi, I can repeat your RTFM on XD-Violence I3d features with very close AP = 0.764 (not your 0.778 yet). However I cannot repeat Sultani on XD-Violence I3d features....

> > As I can see the project is not supposed to be used for real time anomaly detection cause the model doesn't work with only one 16 frame sample...

Think there's no needing in this. Maybe even operation std will be executed faster on data obtained by mean since std will be use zero subtract operation to find the...

Can you specify accuracy for age and gender given after net training?

> Is there are DecisionLevelMax type model for MobileNetV2? https://github.com/qiuqiangkong/audioset_tagging_cnn/issues/67

Sorry guys I don't remember what exactly I changed but eventually I had used OpenPose in [such manner](https://github.com/chamecall/PullUpCounter/blob/030be5855b5c5669be97610d2b8ee7a14789f624/Preprocessor.py#L48).

you can try both options but the logic is the following: first we compute distance matrix based on iou_distance between object's bounding boxes where the less value is the closer...

I guess it's like that: ``` class MobileNetV1_DecisionLevelMax(nn.Module): def __init__(self, sample_rate, window_size, hop_size, mel_bins, fmin, fmax, classes_num): super(MobileNetV1_DecisionLevelMax, self).__init__() window = 'hann' center = True pad_mode = 'reflect' ref =...