search

matterport / Mask_RCNN

Mask R-CNN for object detection and instance segmentation on Keras and TensorFlow
Other
24.57k stars 11.69k forks source link

to train custom dataset not annotated using VIA on mask_RCNN #1866

Open SuganyaMurugesan opened 4 years ago

SuganyaMurugesan commented 4 years ago

Hi All,

I am trying to train Mask_RCNN to detect traffic light. I am using Bosch dataset. The data is not annotated using VIA tool. Please find the sample JSON format of the annotated images,

{ "path": "./rgb/additional/2015-10-05-10-52-01_bag/24594.png", "boxes": [] }, { "path": "./rgb/additional/2015-10-05-10-52-01_bag/24664.png", "boxes": [] }, { "path": "./rgb/additional/2015-10-05-10-52-01_bag/24734.png", "boxes": [] }, { "path": "./rgb/additional/2015-10-05-10-55-33_bag/56988.png", "boxes": [ { "y_min": 278.7432988131, "x_max": 527.1012443072, "occluded": false, "x_min": 518.7379907959, "y_max": 296.3838188664, "label": "Yellow" } ] } Can you please help out in how to accomodate this dataset to load_mask.

basirsedighi commented 4 years ago

you have to check it out but i think that this code dont work unless you use poligons for annotation, which this not

marcfielding1 commented 4 years ago

Have a look under samples and look at the balloon example, the class is designed so it's data format agnostic, https://engineering.matterport.com/splash-of-color-instance-segmentation-with-mask-r-cnn-and-tensorflow-7c761e238b46

I think because it returns both masks and boxes you don't HAVE to have masks.

Check this out https://github.com/heng2j/Mask_RCNN that chap has a model with traffic light detection.