In my opinion, ONNX is more complex than necessary. Therefore, I opted to convert it to an intermediate representation (IR) first, which is then used to generate source code. A key advantage of this approach is the ease of merging nodes into corresponding operations, since ONNX and Burn don't share the same set of operators.
Actually WONNX also transforms to an IR first (early versions did not and simply translated the graph 1:1 to GPU shader invocations in topographically sorted order of the graph). In WONNX the IR nodes are (initially) simply (copy-on-write references to) the ONNX nodes. This IR is then optimized in various ways, including the fusion of ONNX ops (e.g. Conv+ReLU->ConvReLU). The newly inserted node still embeds an ONNX node structure to describe it but uses an internal operator.
I sent you an email a few weeks back - would be great to chat!
WONNX is a seriously impressive project. There is a few reason I didn't just contribute back to WONNX:
1. WONNX does not parse the ONNX model into an IR, which I think is essential to have the freedom to transform the model as required.
2. When I started, WONNX didn't seem focused on symbolic dimensions (but I've seen you shipping the shape inference recently!).
3. The code quality has to be much higher when it's open source! I wanted to hack on this without anyone to please but myself.