LiSsHhUuAaIi opened a new issue, #18475:
URL: https://github.com/apache/tvm/issues/18475

   ### Description
   When converting a PyTorch model containing Spatial Transformer Network (STN) 
operations to TVM Relax module via `torch.export`, an AssertionError occurs. 
TVM currently does not support the `affine_grid_generator.default` and 
`grid_sampler.default` operations that are essential for STNs.
   
   ### Expected behavior
   The PyTorch model with STN operations should be successfully converted to 
TVM Relax module, enabling deployment of spatial transformation models on 
various hardware targets.
   
   ### Actual behavior
   An AssertionError occurs during `from_exported_program` conversion with the 
message `Unsupported function types ['affine_grid_generator.default', 
'grid_sampler.default']`, indicating that TVM's PyTorch frontend lacks support 
for these spatial transformation operations.
   ```
   AssertionError: Unsupported function types ['affine_grid_generator.default', 
'grid_sampler.default']
   ```
   ### Environment
   
   - OS: Ubuntu 20.04.6 LTS
   - TVM version: 0.23.dev0
   - Python version: 3.11.14
   
   ### Steps to reproduce
   
   ```python
   import torch
   import torch.nn as nn
   import torch.nn.functional as F
   import tvm
   from tvm import relax
   
   class MinimalSTNModel(nn.Module):
       def __init__(self):
           super(MinimalSTNModel, self).__init__()
           self.localizer = nn.Sequential(
               nn.Conv2d(3, 8, kernel_size=3),
               nn.ReLU(True),
               nn.AdaptiveAvgPool2d((1, 1)),
               nn.Flatten(),
               nn.Linear(8, 6)
           )
   
       def forward(self, x):
           theta = self.localizer(x)
           theta = theta.view(-1, 2, 3)
           
           # Unsupported operations
           grid = F.affine_grid(theta, x.size())  # 
affine_grid_generator.default
           x = F.grid_sample(x, grid)             # grid_sampler.default
           
           return x
   
   model = MinimalSTNModel()
   model.eval()
   
   x = torch.randn(1, 3, 32, 32)
   
   # PyTorch execution works
   with torch.no_grad():
       output = model(x)
   
   # PyTorch export works  
   exported_program = torch.export.export(model, (x,))
   
   # TVM conversion fails
   from tvm.relax.frontend.torch import from_exported_program
   mod = from_exported_program(exported_program)  # AssertionError here
   ```
   
   ### Triage
   
   * needs-triage
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to