search

visual-layer / fastdup

fastdup is a powerful free tool designed to rapidly extract valuable insights from your image & video datasets. Assisting you to increase your dataset images & labels quality and reduce your data operations costs at an unparalleled scale.
Other
1.54k stars 74 forks source link

[Bug]: Error when switching between regular feature vectors and dinov2 feature vectors #192

Closed dbickson closed 1 year ago

dbickson commented 1 year ago

What happened?

2023-05-15 06:37:41.439 Python[4815:662086] 2023-05-15 05:37:41.439772 [E:onnxruntime:, sequential_executor.cc:368 Execute] Non-zero status code returned while running Squeeze node. Name:'V3Small_feat/global_average_pooling2d_12/Mean_Squeeze__4424' Status Message: /Users/dannybickson/mount/data/onnxruntime/onnxruntime/core/framework/execution_frame.cc:161 onnxruntime::common::Status onnxruntime::IExecutionFrame::GetOrCreateNodeOutputMLValue(const int, int, const onnxruntime::TensorShape , OrtValue &, const onnxruntime::Node &) shape && tensor.Shape() == *shape was false. OrtValue shape verification failed. Current shape:{1,384} Requested shape:{1,576}

What did you expect to see?

Seems like feature vec width is not sent properly

What version of fastdup were you runnning on?

0.929

What version of Python were you running on?

Python 3.8

Operating System

macos, ubuntu

Reproduction steps

import fastdup
fd = fastdup.create(input_dir='hilitu', work_dir='out1111')
fd.run(overwrite=True)
fd.run(overwrite=True, model_path='dinov2s')
fd.run(overwrite=True)

Relevant log output

No response

Attach a screenshot [Optional]

No response

Contact Details [Optional]

No response

dbickson commented 1 year ago

Fixed in 1.13