How to fine-tune SlowFast Kinetics model

[K3N60]

Hi, thank you for sharing the great code.

We try to fine-tune SlowFast Kinetics model with UCF101 dataset, could you provide fine-tune code? Thank you very much.

[97-tianyan]

Hello, I want to train my own data set, which contains 19 behaviors. I want to start from training. I plan to adjust the entire network in the future. Can I send me a script for fine-tuning the code? I will be grateful. my email address is zhuzhutianhao@163.com

[97-tianyan]

嗨，谢谢您的跟进！我假设您已经实现了数据加载器，并且只要求提供微调脚本。我已经准备好脚本，请在下面分享您的电子邮件地址，然后我将把yaml文件发送给您。

非常感谢您，我的电子邮件：oliviald@126.com

Hello, could you please share the script of fine-tuning with me? My email is zhuzhutianhao@163.com

[mrTahion]

Hello @takatosp1, thanks for awesome codebase! I would like to play with both i3d and slowfast(mb try multigrid training) using my own dataset. If it wouldn't be hard for you, could you please, send me fine-tuning script. My email: naulalica@gmail.com. Thank you very much!

[sdjsngs]

every one i have run this code on HMDB51 andUCF101 for some days ， the data loader code you can use in https://github.com/r1ch88/SlowFastNetworks , there are some video are less than 64 frames , so if you pass this video the finetune test top-1 Accuracy is much higher， this is a wrong way , i made interpolation . On HMDB51 i got 76/78/76 test top-1 accuracy on there splits. and they do not send me code so you guys better do it by yourself

[yangxuerong]

Hi! @takatosp1 I also want to finetune slowfast at my own dataset, which has the similar format with Kinetics . I plan to finetune the last FClayer,just few classes. Could you kindly send me a fine-tune code? My mail is: "1710452713@qq.com".

[francescoragusa]

Hi @takatosp1 , I need the finetuning script to finetune Slowfast at my own dataset which is similar to Kinetics. I want to finetune last FC layers at this moment, but in the future I could finetune all the network. Please, can you share the finetuning script? My email is: francesco.ragusa@unict.it. Thank you!

[doursand]

hello, I m interested as well. Looking at the amount of ppl requesting the same thing here wouldn't it be wiser to post a snippet in this thread ?

[Mihuadong]

Hello, I want to train my own data set, which contains 21 behaviors. I want to start from training. I plan to adjust the entire network in the future. Can I send me a script for fine-tuning the code? I will be grateful. my email address is mhd971211@gmail.com

[Peratham]

https://gist.github.com/Peratham/83c02a030a7cd81a1abd644b72dc984e

'''
Here's the snippet from my code.
Also modify the .yaml config file as needed.
Hope this helps! 
'''

from slowfast.models import model_builder, head_helper
from slowfast.config.defaults import get_cfg
import slowfast.utils.checkpoint as cu 
# slowfast net
cfg = get_cfg()
np.random.seed(cfg.RNG_SEED)
torch.manual_seed(cfg.RNG_SEED)
# Load config from cfg.
cfg_file = 'configs/Kinetics/SLOWFAST_NLN_8x8_R50.yaml'
cfg.merge_from_file(cfg_file)
self.slowfast_model = model_builder.build_model(cfg)
cfg.TRAIN.CHECKPOINT_TYPE = 'caffe2'
cfg.TRAIN.CHECKPOINT_FILE_PATH = 'models/SLOWFAST_8x8_R50.pkl'
cu.load_checkpoint(
cfg.TRAIN.CHECKPOINT_FILE_PATH,
  self.slowfast_model,
  False,
  None,
  inflation=False,
  convert_from_caffe2=cfg.TRAIN.CHECKPOINT_TYPE == "caffe2",
)

[Peratham]

@francescoragusa told me (offline/DM) that the finetuning scripts are already uploaded in the `tools' folder.

[yosagaf]

@Peratham i checked the tools folder and I didn't see any finetuning script.

[Peratham]

@Peratham i checked the tools folder and I didn't see any finetuning script.

Modify the code a little bit. Start from the train_net.py and add some model loading to resume from a checkpoint. Finetuning = train from checkpoint with a new last layer or more.

[xiake007]

Hi @takatosp1

Could you share the fine-tune script with me as well? I will train UCF101，my email is zhuwenhe001@163.com

[ciwei123]

Hi! @takatosp1 I am trying to train the model on a custom dataset with 7 labelled actions, using the AVA checkpoint. From previous comments, I understood that just modifying the MODEL.NUM_CLASSES parameter would be enough, but then I get this error: RuntimeError: Error(s) in loading state_dict for SlowFastModel: size mismatch for head.projection.weight: copying a param with shape torch.Size([80, 2304]) from checkpoint, the shape in current model is torch.Size([7, 2304]). size mismatch for head.projection.bias: copying a param with shape torch.Size([80]) from checkpoint, the shape in current model is torch.Size([7]). So I understand that it is not just by doing that, or maybe I have to modify something else. Anyone can give me a clue on this? In case there is no alternative than the finetune network, may you send it to me as well? Email: santiago.herrero@estudiante.uam.es Thank you very much! @SantiHM23 I encountered the same problem, could you tell me the solution?Thank you very much!!

[ciwei123]

@97-tianyan Could you send me a script for fine-tuning the code? Thank you very much!! 2606789239@qq.com

[habatua]

Hello, I want to train my own data set, which contains 21 behaviors. I want to start from training. I plan to adjust the entire network in the future. Can I send me a script for fine-tuning the code? I will be grateful. my email address is mhd971211@gmail.com

Hi，37/5000
Have you received the fine-tuning script? Can you send me a copy? 769370796 @qq.com.Thanks

[bdebbabi]

Hello @takatosp1, I also would like to fine-tune the SlowFast model on a custom dataset. Could you send me the script on this email: debbabi.bilel@gmail.com ? Thank you in advance

[JoeQvQ]

@takatosp1 Thank you fo your great work! can I get the finetune script for UCF101 or custom datasets? Could you please send the script to jj834125468@gmail.com? Thank you very much!

[mikeyEcology]

@takatosp1 Is it possible for me to also get the script to finetune UCF101 for custom datasets? (mtabak [at] west-ulc.ca). Also, is there a script to train a model from scratch using SlowFast with a custom dataset?

[o5786486]

Hi @takatosp1

I would like to request the finetuning script too.

Thank you. o5739201@gmail.com

[rogercmq]

Hi @takatosp1

Thanks for your excellent work! I would like to request the finetuning script too. My email is rogercmq@163.com.

Regards.

[denis19973]

Hi @takatosp1

Thanks for your work! I would like to request the finetuning script.

Thank you. oxingteam@gmail.com

[jessychou]

Hi @Peratham

Thank you for your great work! I'd like to request the finetuning script, it'll be even better if there is one for AVA model. If not, could you please share the existing one. Thanks. zhoufen24@gmail.com

[asafDahan]

Hi @takatosp1. Thanks for your excellent work. Could you share with me your finetuning script? My email is easafdahan@gmail.com

Thanks!

[adityac-deepnorth]

Hi, Anyone who has received the fine tuning script, could you please send it to me as well. It'll be much appreciated. email: adityac@bu.edu

Thanks in advance!

[LevShechter]

Hi @takatosp1

Thanks for your help, I would like to request the finetuning script as well in order to train on my own dataset. My email is levshechter@gmail.com

Thanks!

[ishaan-shah]

Hi @takatosp1 @haooooooqi ,

Could you share the finetuning script with me as well? I want to use the script with the UCF101 dataset. My email is ishaan_shah@alumni.brown.edu. Thank you for your help!

[Heedeok]

Hi @takatosp1 ,

I would like to request the finetuning script too.

Thank you.

Would you share the scripts? gmlejr456@gamil.com

[hansdom]

Hi, Thanks everyone for using the codebase! The finetune script I shared with @oLIVIa-Ld is for her own task. Could everyone reply here to let me know what is the task (dataset) you are working on, and how would you want to perform "finetune" (only tune a fc layer, or tune the entire network), then I can commit an universal finetune script.

@ElainW @anhminh3105 @batmanfightpredator @K3N60

Hi, Thanks everyone for using the codebase! The finetune script I shared with @oLIVIa-Ld is for her own task. Could everyone reply here to let me know what is the task (dataset) you are working on, and how would you want to perform "finetune" (only tune a fc layer, or tune the entire network), then I can commit an universal finetune script.

@ElainW @anhminh3105 @batmanfightpredator @K3N60

Can you send me a copy? My email address is 3532587680@qq.com

[Abiems]

Hi @takatosp1 & @oLIVIa-Ld

Thank you very much for your work! I would like to request the finetuning script.

my email : emsilkh393@gmail.com

[kangsive]

Hi, Thanks everyone for using the codebase! The finetune script I shared with @Olivia-Ld is for her own task. Could everyone reply here to let me know what is the task (dataset) you are working on, and how would you want to perform "finetune" (only tune a fc layer, or tune the entire network), then I can commit an universal finetune script.

@ElainW @anhminh3105 @batmanfightpredator @K3N60

Hi @oLIVIa-Ld , does the fine-tune script work on your task? I would also like to have the script to perform enhancement task, could you kindly please share the script with me? my email address is: kangsive@yahoo.com, thank you!

[namanbiyani]

Hi @takatosp1 & @oLIVIa-Ld & @haooooooqi

Thank you very much this useful work! I would like to request the finetuning script for training on smaller datasets.

my email : namanbiyani4u@gmail.com

[mbilalshaikh]

Hi, Can you provide the finetune script for ucf101? my email is mbs.techy@gmail.com.

[mbilalshaikh]

I really appreciate your quality work. My experiments for ucf101 with slowfastnet are currently under execution. Please provide fine-tuning script as per your ease.

If you commit a script with fine-tuning for all layers, and just the last few layers would be very helpful for everyone.

Email- mbs.techy@gmail.com Thanks,

[tanveer-sayyed]

Hi @takatosp1,

I need the fine-tuning script for my custom dataset. (Why not release it publicly if so many people need it?)

Please email me the script at tanveer2407@gmail.com

[SwapnilDreams100]

Please send me the finetune script at sp6646@nyu.edu

[namhyeongwoo]

Hi @takatosp1, I'm trying to classify my own dataset which contains 6 classes. Could you send me a copy of the code? My email is csc0208@naver.com thx!

[barrybarrych]

Hi @takatosp1 & @haooooooqi & @oLIVIa-Ld I want to fine-tune the slowfast Could you send me a copy of the code? This is my e-mail barrybarrych@gmail.com Thanks!

[DoranLyong]

Hello @takatosp1 & @haooooooqi & @oLIVIa-Ld I want to fine-tune the slowfast. Could you send me a copy of the code either?

My e-mail cheeryun@gmail.com

Have a nice day :)

[ybkim95]

Hi @DoranLyong @barrybarrych @namhyeongwoo I want to fine-tune the slowfast Could you send me a copy of the code? This is my e-mail ybkim95.ai@gmail.com Thanks!

[Balakishan77]

Hi @takatosp1 & @haooooooqi & @oLIVIa-Ld

I want to finetune the entire slowfast network on custom AVA format dataset. Could you please share the fine-tuning script for custom dataset? Please send to my email id: balakishan6@gmail.com

@haooooooqi If you could add this as enhancement to slowfast, it would be really helpful.

Thanks!

[daonguyenkite]

Hi @takatosp1 & @oLIVIa-Ld & @haooooooqi

Thank you very much for the work! I would like to request the finetuning script for my custom dataset with some labels in the ava dataset.

Please send me a copy of script via my email: anhdao1411.work@gmail.com Thanks!

[alpargun]

Hi, I am training using self supervised learning on Berkeley DeepDrive (BDD) dataset which I introduced myself. Then, I need to finetune on a custom driving data set. Could you please send me the finetuning script? Thank you.

argun@ice.rwth-aachen.de

[AndrewCogi]

Hi @DoranLyong @barrybarrych @namhyeongwoo I want to fine-tune the slowfast Could you send me a copy of the code? This is my e-mail [whgustlr0326@gmail.com] Thanks!

[PonyM233]

Hello @takatosp1 @DoranLyong @barrybarrych @namhyeongwoo ，I would also like to fine-tune the SlowFast Kinetics model on the UCF-101 dataset. Could you send me the script? This is my e-mail [innma1020@gmail.com] Thank you

[5baaaaam]

I want to fine-tune it with my own dataset for specific behaviors. Could you send me the script? my e-mail adress is [ obk2287@naver.com] thank you !

[haojc]

Hi, @haooooooqi. I want to finetune the slowfast models on some animal behavior data. And I want to attempt to tune both only a fc layer and the entire network. Could you send me the script?
My email is haojiachang@cibr.ac.cn. Thanks!

[jhyeokj]

Hi, @haooooooqi. I want to finetune the slowfast models on my custom data(anomaly action recognition). Could you please send me the script? This is my e-mail [jhyeok.jeong@gmail.com] Thanks!

[lyj7518]

Hi, @haooooooqi. I'm trying to finetune the slowfast models on UCF101 data. Could you send me the script? This is my e-mail [lyj7518@gmail.com] Thanks!

[StefanoCecconello]

This is my e-mail [ceck1991@gmail.com]. Could you please send me the script? Thanks a lot!