Corrected number of classes

Author: morawi

load_test_maskrcnn.py

@@ -9,6 +9,7 @@
 from models import get_model_instance_segmentation
 from misc_utils import get_dataloaders, sample_images
 import torch
+from draw_image import instance_segmentation_api
 # from PIL import Image
 import argparse
 
@@ -19,26 +20,33 @@
 opt = parser.parse_args()
 
 opt.model_name='maskrcnn_250.pth'
-opt.path2model= 'C:/MyPrograms/t-HPC-results/mask_rcnn-Mar-30-at-3-18/'
+#opt.path2model= 'C:/MyPrograms/t-HPC-results/mask_rcnn-Mar-30-at-3-18/'
+# opt.path2model ='C:/MyPrograms/t-HPC-results/mrcn_no_bkgrnd/'
+opt.path2model ='C:/MyPrograms/t-HPC-results/mrcnn_prsn_detect/'
 opt.HPC_run = 0
-opt.remove_background = 1
-opt.person_detection = 0
+opt.remove_background = False
+opt.person_detection = True
 opt.train_percentage = 0.5
 opt.batch_size = 1
-opt.train_shuffle = 0
+opt.train_shuffle = False
 opt.n_cpu=0
 opt.cuda = False # this will definetly work on the cpu if it is false
 
+folder_name = '/my'
+image_name = 'myimage'
+
+image_path = folder_name+image_name
+
+
 device = torch.device('cuda' if opt.cuda else 'cpu')
-model = get_model_instance_segmentation( number_of_classes(opt.dataset_name) ) 
+
+model = get_model_instance_segmentation( number_of_classes(opt) ) 
 print("loading model", opt.model_name ) 
 model.load_state_dict(torch.load(opt.path2model+opt.model_name, map_location=device )) 
 model.eval()
-data_loader, data_loader_test = get_dataloaders(opt)
+instance_segmentation_api(model, folder_name+image_name)
+
 
-for i in range(3): # range( len(data_loader_test)): # let's just check a couple of images
- images, targets = next(iter(data_loader_test)) # get image(s)
- sample_images(images, targets, model, device)
 
 
 

misc_utils.py

@@ -13,26 +13,28 @@
 import utils
 
 
-def sample_images(images, targets, model, device): 
+def sample_images(images, targets, model, device, number_of_classes): 
  # Image.fromarray(255*targets[0]['masks'].squeeze(0).numpy()).show()
  images = list(image.to(device) for image in images)
  model.eval() # setting model to evaluation mode
  with torch.no_grad():
  predictions = model(images) # Returns predictions
  masks = predictions[0]['masks'].cpu().squeeze(1)
  labels = predictions[0]['labels'].cpu()
+ scores = predictions[0]['scores'].cpu() # scores are already sorted
  model.train() # putting back the model into train status/mode 
 
 
  print(labels)
  print(targets[0]['labels'])
  print('-------------####-----------')
 
- # for i in range(len(labels)): # we have one label for each mask
- # Image.fromarray( 255*masks[i].numpy().round() ).show()
- # print(labels[i])
- # to_pil = transforms.ToPILImage()
- # to_pil(images[0]).show()
+ for i in range(number_of_classes-1): # we have one label for each mask
+ Image.fromarray( 255*masks[i].numpy().round() ).show()
+ print('label:', labels[i], ', score:', scores[i])
+ to_pil = transforms.ToPILImage()
+ to_pil(images[0].cpu()).show()
+ 
 
 def get_dataloaders(opt):
  # Configure dataloaders