Deeplabcut: Train Starting from Checkpoint resets to zero [question]

[phcao101]

Your Operating system and DeepLabCut version
Please state your operating system, env, and which version of DeepLabCut you are using.
Example: Ubuntu 16.04 LTS, with an Anaconda Env, & DeepLabCut1.x or 2.x.

Please complete the following information about your system:

OS: Windows 7
DeepLabCut Version 2.0.7.2
Browser: Chrome

Describe the problem
I've been trying to restart the training at a specific checkpoint but when I begin training the network, it just starts back from 0.

I have followed instructions in regards to changing init_weights.

init_weights: C:UsersAlitoverhead3-Paul_Cao-2019-07-25dlc-modelsiteration-0overhead3Jul25-trainset95shuffle1trainsnapshot-750

Is there anything else I need to do in order to retrain starting at a checkpoint?

Screenshots

pose_cfg.txt

[MMathisLab]

yes, it resets to 0, but if you loaded the weights, it will be using them! (it also states this in our user-guide and documentation)

https://camo.githubusercontent.com/663a60c8ec32c35be321b4c827fe415a839a4e05/68747470733a2f2f737461746963312e73717561726573706163652e636f6d2f7374617469632f3537663664353163396637343536366635356563663237312f742f3563343065666265626261323233373330383832633264652f313534373735393635363332342f426f78322d30312e706e673f666f726d61743d3130303077

also, if you are using 101, you should set intermediate_supervision to True

[phcao101]

Thanks so much!!

[MMathisLab]

no problem. if you like the package, would consider giving us a star :)

[phcao101]

One last question: If doing this enables the network to retrain from a loaded weight, why does the loss still start off so high?