Closed Ekta246 closed 3 years ago
rwightman
commented
3 years ago You are using a custom dataset, I cannot help. Please stop creating issues about this.
I did test a one class scenario w/ TFW validator. After only two epochs of training VOC with all but bicyle filtered
PascalBoxes_Precision/mAP@0.5IOU: 0.6753591421661145
PascalBoxes_PerformanceByCategory/AP@0.5IOU/bicycle: 0.6753591421661145If you do want help with your custom datasets / applications I charge 225 USD / hr for that sort of work. I canot provide assistance otherwise. I only support fixing bugs in the models here w/ provided open datasets and impl that I have verified myself.
Ekta246
commented
3 years ago I hope you have got me wrong. I don't say that it's a bug. I am just seeking help from the other audience as well if they have reached the same issue. Sorry for the misunderstanding.
About the last post, it was labeled as bug by mistake. I am just askinh the possible reasons to do so.
I am sorry to have wasted your time and your inconvenience.
On Thu, Nov 12, 2020, 3:22 PM Ross Wightman notifications@github.com wrote:
If you do want help with your custom datasets / applications I charge 225 USD / hr for that sort of work. I canot provide assistance otherwise. I only support fixing bugs in the models here w/ provided open datasets and impl that I have verified myself.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/rwightman/efficientdet-pytorch/issues/122#issuecomment-726319627, or unsubscribe https://github.com/notifications/unsubscribe-auth/AMQBWSNDX7D4OCOG5EY7WWLSPQ7XLANCNFSM4TTW3KTQ .
rwightman
commented
3 years ago No worries. I'm really trying to keep the issues down to just bugs and feature requests right now. It consumes too much time weeding through valid issues (for me to fix) vs requests for help, etc.
For community help / discussions I'm hoping to activate github Discussions beta for this repo very soon, it doesn't quite qualify yet.
On Thu, Nov 12, 2020 at 12:46 PM Ekta P Bhojwani notifications@github.com wrote:
I hope you have got me wrong. I don't say that it's a bug. I am just seeking help from the other audience as well if they have reached the same issue. Sorry for the misunderstanding.
About the last post, it was labeled as bug by mistake. I am just askinh the possible reasons to do so.
I am sorry to have wasted your time and your inconvenience.
On Thu, Nov 12, 2020, 3:22 PM Ross Wightman notifications@github.com wrote:
If you do want help with your custom datasets / applications I charge 225 USD / hr for that sort of work. I canot provide assistance otherwise. I only support fixing bugs in the models here w/ provided open datasets and impl that I have verified myself.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub < https://github.com/rwightman/efficientdet-pytorch/issues/122#issuecomment-726319627 , or unsubscribe < https://github.com/notifications/unsubscribe-auth/AMQBWSNDX7D4OCOG5EY7WWLSPQ7XLANCNFSM4TTW3KTQ
.
— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub https://github.com/rwightman/efficientdet-pytorch/issues/122#issuecomment-726332556, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABLQICF27A5PX54NL2EJT2LSPRCQZANCNFSM4TTW3KTQ .
Ekta246
commented
3 years ago Yes, I am really sorry.
On Thu, Nov 12, 2020, 4:13 PM Ross Wightman notifications@github.com wrote:
No worries. I'm really trying to keep the issues down to just bugs and feature requests right now. It consumes too much time weeding through valid issues (for me to fix) vs requests for help, etc.
For community help / discussions I'm hoping to activate github Discussions beta for this repo very soon, it doesn't quite qualify yet.
On Thu, Nov 12, 2020 at 12:46 PM Ekta P Bhojwani <notifications@github.com
wrote:
I hope you have got me wrong. I don't say that it's a bug. I am just seeking help from the other audience as well if they have reached the same issue. Sorry for the misunderstanding.
About the last post, it was labeled as bug by mistake. I am just askinh the possible reasons to do so.
I am sorry to have wasted your time and your inconvenience.
On Thu, Nov 12, 2020, 3:22 PM Ross Wightman notifications@github.com wrote:
If you do want help with your custom datasets / applications I charge 225 USD / hr for that sort of work. I canot provide assistance otherwise. I only support fixing bugs in the models here w/ provided open datasets and impl that I have verified myself.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <
https://github.com/rwightman/efficientdet-pytorch/issues/122#issuecomment-726319627
, or unsubscribe <
https://github.com/notifications/unsubscribe-auth/AMQBWSNDX7D4OCOG5EY7WWLSPQ7XLANCNFSM4TTW3KTQ
.
— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub < https://github.com/rwightman/efficientdet-pytorch/issues/122#issuecomment-726332556 , or unsubscribe < https://github.com/notifications/unsubscribe-auth/ABLQICF27A5PX54NL2EJT2LSPRCQZANCNFSM4TTW3KTQ
.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/rwightman/efficientdet-pytorch/issues/122#issuecomment-726345516, or unsubscribe https://github.com/notifications/unsubscribe-auth/AMQBWSLQB4GQR47CM4BK263SPRFY3ANCNFSM4TTW3KTQ .
Is your feature request related to a problem? Please describe. A clear and concise description of what the problem is. mAP always 0 after validating on every epoch
I am training the model efficientdet-d0 based on pretrained Coco weights from the repo config file on a custom dataset of 1 class only and 1 object per image. I am just resetting the coco head of 90 class to 1 class. Rest of the part trains as it is. There is no freezing of layers Describe the solution you'd like A clear and concise description of what you want to happen. sense of mAP Describe alternatives you've considered A clear and concise description of any alternative solutions or features you've considered. Do I need to consider freezing of the layers?
Additional context Add any other context or screenshots about the feature request here.