operands could not be broadcast together with shapes (8,256) (4,256,256)

[feivelliu]

hello，When I run,：python train.py --dataset pcontext --model deeplab --aux --backbone resnest200 I found the following error：

  File "/dahuafs/userdata/229288/00_deeplearning/01_pytorch/segEncoding/experiments/segmentation/train.py", line 238, in validation
    correct, labeled, inter, union = eval_batch(self.model, image, target)
  File "/dahuafs/userdata/229288/00_deeplearning/01_pytorch/segEncoding/experiments/segmentation/train.py", line 229, in eval_batch
    inter, union = utils.batch_intersection_union(pred.data, target, self.nclass)
  File "/dahuafs/userdata/229288/00_deeplearning/anaconda3/envs/scseg/lib/python3.7/site-packages/torch_encoding-1.2.2b20201023-py3.7-linux-x86_64.egg/encoding/utils/metrics.py", line 122, in batch_intersection_union
    predict = predict * (target > 0).astype(predict.dtype)
ValueError: operands could not be broadcast together with shapes (8,256) (4,256,256) 

        def eval_batch(model, image, target):
            outputs = model(image)
            outputs = gather(outputs, 0, dim=0)
            pred = outputs[0]
            target = target.cuda()
            correct, labeled = utils.batch_pix_accuracy(pred.data, target)
            inter, union = utils.batch_intersection_union(pred.data, target, self.nclass)
            return correct, labeled, inter, union

Is there something wrong with this code？

[khanhhungvu1508]

Hello, I am also facing this problem with dataset ade20k, backbone resnet50 and model encnet.

Cell In[64], line 121, in Trainer.validation(self, epoch)
    119 for i, (image, target) in enumerate(tbar):
    120     with torch.no_grad():
--> 121         correct, labeled, inter, union = eval_batch(self.model, image, target)
    123     total_correct += correct
    124     total_label += labeled

Cell In[64], line 112, in Trainer.validation.<locals>.eval_batch(model, image, target)
    110 print(f"target.shape = {target.shape}")
    111 correct, labeled = batch_pix_accuracy(pred.data, target)
--> 112 inter, union = batch_intersection_union(pred.data, target, self.nclass)
    113 return correct, labeled, inter, union

Cell In[59], line 104, in batch_intersection_union(output, target, nclass)
    101 predict = predict.cpu().numpy().astype('int64') + 1
    102 target = target.cpu().numpy().astype('int64') + 1
--> 104 predict = predict * (target > 0).astype(predict.dtype)
    105 intersection = predict * (predict == target)
    106 # areas of intersection and union

ValueError: operands could not be broadcast together with shapes (150,240) (16,240,240) 

How to solve them?

[zhanghang1989]

This is a really old and unmaintained repo. Please use more recent projects, e.g. mmsegmentation