search

HRNet / HRNet-Object-Detection

Object detection with multi-level representations generated from deep high-resolution representation learning (HRNetV2h). This is an official implementation for our TPAMI paper "Deep High-Resolution Representation Learning for Visual Recognition". https://arxiv.org/abs/1908.07919
Apache License 2.0
644 stars 97 forks source link

Results of cascade_rcnn_hrnetv2p_w32 #5

Closed swordlidev closed 5 years ago

swordlidev commented 5 years ago

Hi @leoxiaobin When I use the cascade_rcnn_hrnetv2p_w32 model for inference on val2017. I got 0.389, instead of 43.7mAP. Why？

wondervictor commented 5 years ago

@lijiannuist I'll check it now.

wondervictor commented 5 years ago

@lijiannuist I've tested Cascade R-CNN on val2017 just and obtained the results below: image

Can you provide more information about your modifications or how you migrate this code into your environment as well as data preprocessing.

wondervictor commented 5 years ago
  1. have you modified configurations for data in cascade_rcnn_hrnetv2p_w32.py?
  2. do you use the downloaded model file ?
wondervictor commented 5 years ago

Can you provide part of your configurations about dataset.

wondervictor commented 5 years ago

I just change the data format, and did not use zip data. because CocoZipDataset has bug. image

Images are not found in your zip file. not bug

swordlidev commented 5 years ago

image

when I unzip zip file and use CocoDataset, it works and images are found.

wondervictor commented 5 years ago

You need to set to_rgb=True because CocoDataset default format is BGR. You can refer to the comments below

swordlidev commented 5 years ago

Hi @wondervictor It works when to_rgb is set to be True

wondervictor commented 5 years ago

solved. I'll close it.