search

matterport / Mask_RCNN

Mask R-CNN for object detection and instance segmentation on Keras and TensorFlow
Other
24.63k stars 11.7k forks source link

Getting Mask Pixel Coordinate #365

Open tsly123 opened 6 years ago

tsly123 commented 6 years ago

Hi,

I am new to deep learning. Could you give me some guideline how to:

  1. get the pixel coordinate of the mask (4 greatest values, corresponding to 4 directions of the image) and the bounding boxes.
  2. I try to locate only the person in image (dataset UCF-101). Do I have to fine tune with my dataset first?

Thank you.

derelearnro commented 6 years ago

I am the same project. I need to return the center of mass of the multiple objects. and find the orientation of them.

tsly123 commented 6 years ago

I ran the demo.ipynb with the images from dataset UCF-101 (320x240x3) with 81 classes. Although there are clearly persons on the image, the demo cannot recognize it. There are papers using this dataset (300x300x3) on Faster R-CNN and the results are good.

Does the quality of the image affect the result? Cause' as I understand, Faster R-CNN and Mask R-CNN have the same detecting engine.

2 images result from same demo.ipynp file. BaseballDunk (320x240x3) from UCF-101, the other one (640x415x3)is available in the repo. This baseballDunk result is better from some other similar baseballDunk images of which the result is a big boat, instead of persons. baseballdunk test_image

I guess I have to fine tune the network with my dataset (UCF101) with only 2 classes, BG and person. But I dont have the ground truth fo this dataset? Any solution? Thank you.

iliachigogidze commented 6 years ago

Have you found a solution for your issue? I have the same problem...

everysens-mounsif-hanafi commented 4 years ago

Hello, i have the same struggle, i need to find mask contour pixels coordinates, if someone has some advice, it would be helpful

sohinimallick commented 3 years ago

Hello. Did someone find a solution?