Iteration Plan - Dec 2020

[dreamerlin]

Since MMAction2 v0.9 has been released, now it's time to plan for the next iteration, i.e. v0.10.0.
This will be a one-month iteration and our focus in this iteration is to solve the legacy from the last version and do some harmless refactoring to our repo.

Algorithms:

• FastRCNN on AVA ([Feature] Fast RCNN #351)
  • Integration with MMDet
  • Seek for good practices for AVA spatio-temporal pipeline
• SlowFast on AVA ([Feature] Fast RCNN #351)
• AVSlowFast ([Feature] Add Resnet2dAudio #355)
• TAM ([Feature] Support TAM [WIP] #304)
• SCSampler ([Feature] ACSampler #381)
  • Implement Visual sampler and audio sampler
• MOC ([Feature] Init tubedataset #308, [Feature] Add MOC #326)
  • Construct data pipeline for UCF101-24 and JHMDB
  • Improve the efficiency of tubelet dataset pipeline
  • Implement MOC

Improvement

• Improve performance of AudioOnly
• Refactor our config file structure ([Refactor]Refactor config structure #488)
• Refactor our unittest file structure ([Refactor] Refactor unit test structure #433)

---

Deferred

• MOC ([Feature] Init tubedataset #308, [Feature] Add MOC #326)
  • Construct data pipeline for UCF101-24 and JHMDB
  • Improve the efficiency of tubelet dataset pipeline
  • Implement MOC
• Improve performance of AudioOnly
• Refactor our config file structure ([Refactor]Refactor config structure #488)
• Refactor our unittest file structure ([Refactor] Refactor unit test structure #433)
• TAM ([Feature] Support TAM [WIP] #304)
• SCSampler ([Feature] ACSampler #381)
  • Implement Visual sampler and audio sampler

[irvingzhang0512]

Plan to implement

• TSM-MobileNetV2 ([Feature] Support TSM-MobileNetV2 #415)
• YOWO(After tubedataset is ready)

[wwdok]

Do you have plan to add action localization demo ? It seems if we input video, it can generate proposal file that like caption ?

[innerlee]

@wwdok Feature requests goes ➡️ here #19

[wwdok]

Got it, Thank you !