ONNX error input tensor must have at least 2 dimensions

[wanglong001]

Describe the bug
onnx model of yolov5 to burn error:

ERROR burn_import::logger: PANIC => panicked at /home/wzl/.cargo/git/checkouts/burn-178c6829f420dae1/f43b686/burn-import/src/onnx/op_configuration.rs:229:9:
Flatten: input tensor must have at least 2 dimensions (got 0)    
thread 'main' panicked at /home/wzl/.cargo/git/checkouts/burn-178c6829f420dae1/f43b686/burn-import/src/onnx/op_configuration.rs:229:9:
Flatten: input tensor must have at least 2 dimensions (got 0)

To Reproduce

    std::env::set_var("OUT_DIR", ".");
    let onnx_path_str = "/tmp/slient.onnx";
    // let onnx_path = std::path::PathBuf::from(onnx_path_str);
    burn_import::onnx::ModelGen::new()
        .input("/tmp/best.onnx")
        .out_dir("model/")
        .run_from_script();

Expected behavior
no panic

Screenshots

Desktop (please complete the following information):

• OS: window11 WSL2

[antimora]

I will look into this soon. I have been busy with PyTorch import implementation.

[KarstenB]

I encountered the same issue, and it boils down to the fact that your Split layer, or in my case LeakyReLu is not supported: https://github.com/tracel-ai/burn/blob/main/burn-import/SUPPORTED-ONNX-OPS.md

[polarathene]

Reproduction

For reference, using the current git version of Burn (has some improvements since the latest release tag v0.17.1).

Reproduction uses this compose.yaml.

No specific ONNX model linked, assuming YOLOv5 (v7.0 release): https://github.com/ultralytics/yolov5/releases/tag/v7.0

ONNX_MODEL_URL='https://github.com/ultralytics/yolov5/releases/download/v7.0/yolov5n.onnx' \
docker compose run --rm -it burn-onnx-import cargo build

   INFO onnx_ir::from_onnx: Parsing ONNX file: src/model/example.onnx
  DEBUG onnx_ir::from_onnx: Number of nodes: 263
  DEBUG onnx_ir::from_onnx: Number of inputs: 1
  DEBUG onnx_ir::from_onnx: Number of initializers: 120
  DEBUG onnx_ir::from_onnx: Number of outputs: 1
  DEBUG onnx_ir::from_onnx: Opset domain: "<default>", version: 17
  ERROR burn_import::logger: PANIC => panicked at /root/.cargo/git/checkouts/burn-6c277d792b0d5d7a/1f5ccd4/crates/onnx-ir/src/ir.rs:38:33:
  invalid tensor model.0.conv.weight

[antimora]

I checked, it's failing on a brand new issue:

DEBUG onnx_ir::from_onnx: Adding node "reshape6"
DEBUG onnx_ir::proto_conversion: Converting ONNX node with type "Concat"
DEBUG onnx_ir::from_onnx: Renaming node "/model.24/Concat_3" to "concat17"
DEBUG onnx_ir::rank_inference: Inferring rank for node: concat17
DEBUG onnx_ir::node::concat: Concat rank inference for node concat17
DEBUG onnx_ir::node::concat: Concat using tensor input rank for concat17: 3
DEBUG onnx_ir::node::concat: Concat output rank for concat17: 3
DEBUG onnx_ir::rank_inference: Rank inference result for concat17: [Argument { name: "output0", ty: Tensor(TensorType { elem_type: Float16, rank: 3, static_shape: None }), value: None, passed: false }]
DEBUG onnx_ir::from_onnx: Adding node "concat17"
 INFO onnx_ir::from_onnx: Finished parsing ONNX file: ./out/yolov5n.onnx
DEBUG burn_import::onnx::to_burn: Writing debug onnx graph file: "./out/yolov5n.onnx.txt"
DEBUG burn_import::onnx::to_burn: Writing debug graph file: "./out/yolov5n.graph.txt"
DEBUG burn_import::burn::graph: Registering node => 'conv2d'
ERROR burn_import::logger: PANIC => panicked at crates/burn-import/src/onnx/to_burn.rs:1880:18:
Unsupported tensor type Float16

thread 'main' panicked at crates/burn-import/src/onnx/to_burn.rs:1880:18:
Unsupported tensor type Float16
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
(onnx-tests) [burn-import]%

This probably will be fixed by: #3574

Closing the original issue.