如何阅读TensorFlow非极大值抑制源码？找不到gen_image_ops文件怎么办？

Great question—this is a super common gotcha with TensorFlow's auto-generated files like gen_image_ops! Let me walk you through where to find the actual implementation code:

• Why you can't find gen_image_ops: Files prefixed with gen_ are automatically generated during TensorFlow's build process. They aren't checked into the source code repository directly, which is why you won't spot them when browsing the repo. These files act as a bridge between TensorFlow's low-level C++ ops and the Python API.
• Public Python API wrapper: The user-facing NMS functions (like non_max_suppression()) are defined in tensorflow/python/ops/image_ops.py. This is the file you'd interact with as a user, and it delegates to the auto-generated gen_image_ops under the hood. You'll find documentation and input validation logic here.
• Core C++ implementation: The actual algorithm logic lives in TensorFlow's C++ kernel code. Look for non_max_suppression_ops.cc and non_max_suppression_ops.h in the tensorflow/core/kernels/ directory. This is where the heavy lifting of the NMS algorithm happens—including bounding box overlap calculations and suppression logic.
• Op definition specification: If you want to see how the NMS op is formally defined (input/output types, attributes, etc.), check tensorflow/core/ops/image_ops.pbtxt. This protobuf text file defines the op's interface, which is used to generate the gen_image_ops files during the build process.
• Pro tip for browsing: If you're looking through TensorFlow's source repo, use the built-in search bar to look for "non_max_suppression"—it'll pull up all relevant files instantly, so you don't have to navigate directories manually.

内容的提问来源于stack exchange，提问作者mkocabas