Any technical questions and questions related to model conversion to TFLite / ONNX / etc. formats can be asked in this thread.
Mobile AI Workshop Technical and Model Conversion Questions
- Thread starter Andrey Ignatov
- Start date
farazkayani
New member
Hello,
I attempted to upload the ZIP file (containing the required folders, images generated from the model in root folder, and both TFLite models) for MAI Image Denoising Challenge. All other folders in the ZIP are empty except the TFLite folder. However, the server returned the following errors:
Error occured while parsing readme.txt. Please make sure you have a line for runtime, cpu/gpu, extra data and other (4 lines in total).
Parsed information:
Runtime: -1.000000
CPU/GPU: -1
Data: -1
Other:
Traceback (most recent call last):
File "/app/program/evaluation.py", line 95, in <module>
raise Exception('Expected %d .png images'%len(ref_pngs))
Exception: Expected 50 .png images
What is the issue here?
Thank you very much for your assistance.
I attempted to upload the ZIP file (containing the required folders, images generated from the model in root folder, and both TFLite models) for MAI Image Denoising Challenge. All other folders in the ZIP are empty except the TFLite folder. However, the server returned the following errors:
Error occured while parsing readme.txt. Please make sure you have a line for runtime, cpu/gpu, extra data and other (4 lines in total).
Parsed information:
Runtime: -1.000000
CPU/GPU: -1
Data: -1
Other:
Traceback (most recent call last):
File "/app/program/evaluation.py", line 95, in <module>
raise Exception('Expected %d .png images'%len(ref_pngs))
Exception: Expected 50 .png images
What is the issue here?
Thank you very much for your assistance.
If you are making the final submission - you can ignore all issues.
For the main phase, Codabench is used only for computing numerical (PSNR/SSIM) results on the validation set, thus you need to follow the instructions on the corresponding webpage and upload your reconstructed images there.
Dear Andrey Ignatov,
I am very interested in the Efficient Stable Diffusion on Apple challenge. However, I don't have access to an Apple M4 device myself. I'm wondering if there are any alternative ways to test my model besides submitting to Codabench? I would like to avoid exhausting the 3-submission limit for preliminary testing purposes.
Thank you for your time and for organizing such an inspiring workshop.
Best regards
I am very interested in the Efficient Stable Diffusion on Apple challenge. However, I don't have access to an Apple M4 device myself. I'm wondering if there are any alternative ways to test my model besides submitting to Codabench? I would like to avoid exhausting the 3-submission limit for preliminary testing purposes.
Thank you for your time and for organizing such an inspiring workshop.
Best regards
You can test your model on any machine with MacOS: CoreML GPU delegate supports both Apple and AMD/Nvidia silicon, so if your solutions works - then at least GPU-based inference would be available on the Apple M4.
Hello,
I'm working on the final results submission for the Mobile AI 2026 Real Image Denoising Challenge. The PyTorch-to-TFLite export works correctly for a static input image size and runs well on a mobile device. However, exporting with dynamic input shapes of [1, None, None, 3], as required for the second TFLite model export for the challenge, proves to be problematic.
Would it be acceptable - similar to last year - to submit a TFlite model version with static input shapes for quality evaluation? If so, please let me know the required resolution. Is it the same as the [1, 2432, 3200, 3] images in the validation subset?
Thank you in advance.
I'm working on the final results submission for the Mobile AI 2026 Real Image Denoising Challenge. The PyTorch-to-TFLite export works correctly for a static input image size and runs well on a mobile device. However, exporting with dynamic input shapes of [1, None, None, 3], as required for the second TFLite model export for the challenge, proves to be problematic.
Would it be acceptable - similar to last year - to submit a TFlite model version with static input shapes for quality evaluation? If so, please let me know the required resolution. Is it the same as the [1, 2432, 3200, 3] images in the validation subset?
Thank you in advance.
Yes, it's acceptable, the resolution is the same.