def visualize_epoch(self, model, data_loader, priorbox, writer, epoch, use_gpu): model.eval() img_index = random.randint(0, len(data_loader.dataset)-1) #img_index = 1 # get img image = data_loader.dataset.pull_image(img_index) anno = data_loader.dataset.pull_anno(img_index) # visualize archor box viz_prior_box(writer, priorbox, image, epoch) # get preproc preproc = data_loader.dataset.preproc preproc.add_writer(writer, epoch) # preproc.p = 0.6 # preproc image & visualize preprocess prograss images = Variable(preproc(image, anno)[0].unsqueeze(0), volatile=True) if use_gpu: images = images.cuda() # visualize feature map in base and extras base_out = viz_module_feature_maps(writer, model.base, images, module_name='base', epoch=epoch) extras_out = viz_module_feature_maps(writer, model.extras, base_out, module_name='extras', epoch=epoch) # visualize feature map in feature_extractors viz_feature_maps(writer, model(images, 'feature'), module_name='feature_extractors', epoch=epoch) model.train() images.requires_grad = True images.volatile=False base_out = viz_module_grads(writer, model, model.base, images, images, preproc.means, module_name='base', epoch=epoch)
def visualize_epoch(self, model, data_loader, priorbox, writer, epoch, use_gpu): model.eval() img_index = random.randint(0, len(data_loader.dataset)-1) # get img image = data_loader.dataset.pull_image(img_index) anno = data_loader.dataset.pull_anno(img_index) # get preproc preproc = data_loader.dataset.preproc preproc.add_writer(writer, epoch) # visualize archor box viz_prior_box(writer, priorbox, image, epoch) # preproc image & visualize preprocess prograss images = Variable(preproc(image, anno)[0].unsqueeze(0), volatile=True) if use_gpu: images = images.cuda()