search

bethgelab / siamese-mask-rcnn

Siamese Mask R-CNN model for one-shot instance segmentation
Other
346 stars 60 forks source link

Prediction time during testing #13

Closed trungpham2606 closed 5 years ago

trungpham2606 commented 5 years ago

Hi @mbethge , Thank you for sharing your great work. I just downloaded your pre-trained model and made some tests on my own dataset. The first thing I can see is that the time this model needs to predict 1 image is quite slow, about more than 7s for 1 image. What's about your prediction time for 1 image ?

michaelisc commented 5 years ago

Which model did you download and which function did you use for testing? model.evaluate_dataset or model.detect? The latter one can take a long time for the first image, as it has to initialize some components when the model is called for the first time. I found this behavior to be the same for the original Matterport Mask R-CNN implementation we build upon.

trungpham2606 commented 5 years ago

@michaelisc I downloaded large_siamese_mrcnn_coco_full model and used model.detect for detection. I was testing detection for multiple images. For the first image, it took about 24sec, and approximately 5 to 7sec for each image after that. Ah, I just figured out that for prediction part, the model takes only 0.5 to 0.7 sec for 1 image. But display_results part takes so much time (at least 3sec for 1 image in my PC).

michaelisc commented 5 years ago

Between 500ms and 700ms sounds about right. Did that finding answer your question then? I am not sure if I can do anything to speed up the visualization function. I guess the bottleneck there is matplotlib but never profiled it.

trungpham2606 commented 5 years ago

@michaelisc Iam trying to use opencv instead to see whether it can boost the visualization time or not. I will report the results later. Update: I think the problem comes from my PC, after trying several times again with matplotlib and opencv, the total time results ( prediction + visualization ) are the same. Thank you for your prompt support.