imvision12/segformer-tf

A Tensorflow Implementation of "SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers"

Hi 👋, I'm IMvision

👀 I’m interested in Machine Learning and Deep Learning

• 🎓 I am a Computer Science student, just started my journey in Open-Source.
• 👯 I’m looking forward to collaborate on open source projects.
• 💬 Ask me about computer vision, tensorflow, pytorch
• 🌱 I’m currently learning: NLP, Generative AI(GPT, Stable Diffusion)
• ✔ Fun fact : I Always try to learn something new and then sleep till it store in the brain 😎

🔨💻 Programming Languages, Frameworks, Libraries, Databases and Tools

⚡ Stats ⚡

🏆 GitHub Trophies

segformer-tf's Issues

Which depencies are used?

Dear IMVision12,
I was wondering which dependencies you used to build this project with? I am getting an error while loading the model in the seformer notebook:

ValueError: Attempt to convert a value (None) with an unsupported type (<class 'NoneType'>) to a Tensor.

Exception encountered when calling MixVisionTransformer.call()

Hello, thanks for implementing this model. I'm having the following error when I try to instantiate a SegFormer, following the example in the repository. Any ideas how to fix it?

model = SegFormer_B3(input_shape = (224, 224, 3), num_classes = 19) 

  File D:\marcel\testes_diversos\notebook_atual\segformer_tf\models\segformer.py:93 in SegFormer_B3
    x = MixVisionTransformer(

  File D:\b3\envs\gdal_env\lib\site-packages\keras\src\utils\traceback_utils.py:123 in error_handler
    raise e.with_traceback(filtered_tb) from None

  File D:\marcel\testes_diversos\notebook_atual\segformer_tf\models\modules.py:266 in call
    x = self.call_features(x)

  File D:\marcel\testes_diversos\notebook_atual\segformer_tf\models\modules.py:234 in call_features
    x = blk(x, H, W)

  File D:\marcel\testes_diversos\notebook_atual\segformer_tf\models\modules.py:89 in call
    x = x + self.drop_path(self.attn(self.norm1(x), H, W))

RuntimeError: Exception encountered when calling MixVisionTransformer.call().

Could not automatically infer the output shape / dtype of 'mix_vision_transformer_2' (of type MixVisionTransformer). Either the `MixVisionTransformer.call()` method is incorrect, or you need to implement the `MixVisionTransformer.compute_output_spec() / compute_output_shape()` method. Error encountered:

Exception encountered when calling Layer.call().

Layer Layer does not have a `call()` method implemented.

Arguments received by Layer.call():
  • args=('tf.Tensor(shape=(None, None, 64), dtype=float32)',)
  • kwargs=<class 'inspect._empty'>

Arguments received by MixVisionTransformer.call():
  • args=('<KerasTensor shape=(None, 224, 224, 3), dtype=float32, sparse=None, name=keras_tensor_10>',)
  • kwargs=<class 'inspect._empty'>