Trained culane-small but could not get the performance described in the paper.

[KiveeDong]

I trained culane-small model using CULane train split(88.9K) for 16 epochs, but the F1-score on CULane test split(34.7K) is only 77.55.
The F1-score reported in the paper is 78.14, and the gap is 0.69.
I think this gap is not small, as the F1-score gap between samll model and medium model is only 0.6 .
So, could you kindly tell us some details about how to train the model reported in the paper?
By the way, I use a batch size of 32 and did not change any hyperparameter in the codes.

Thank you very much~

[wangnian97]

同学，想请问下这个是怎么训练呢。CUDA_VISIBLE_DEVICES=0 PORT=29001 tools/dist_train.sh configs/condlanenet/culane/culane_small_train.py 1 --no-validate ，我用这个在单gpu训练，但是报错

[KiveeDong]

同学，想请问下这个是怎么训练呢。CUDA_VISIBLE_DEVICES=0 PORT=29001 tools/dist_train.sh configs/condlanenet/culane/culane_small_train.py 1 --no-validate ，我用这个在单gpu训练，但是报错

直接用tools/train.py训练

[wangnian97]

可以看下你在服务器训练时的具体命令吗

[shuizaola]

me too .how to train

[CongerW]

同学，你这个定量结果F1是怎么得到的啊，求教~

[jyang68sh]

他不是带evaluate 选项么

[KiveeDong]

同学，你这个定量结果F1是怎么得到的啊，求教~

@CongerW 他会保存测试结果，然后用scnn源码里culane的evaluate工具就可以计算指标

[jyang68sh]

@KiveeDong 他test脚本里带evaluate选项

[KiveeDong]

@KiveeDong 他test脚本里带evaluate选项

@jyang68sh 我看只有curvelanes的测试脚本带了evaluate参数，tusimple和culane都没有

[jyang68sh]

@KiveeDong 确实如此，但是他定义了 LaneMetricCore 这个类，调用就行了

[CongerW]

@KiveeDong 确实如此，但是他定义了 LaneMetricCore 这个类，调用就行了

老哥，你CULane定量分析的代码还在吗，可以分享下不，我配置SCNN测试环境配置了半天一直没成功

[jyang68sh]

你可以看下laneatt
他有python的量化代码，但是速度很慢。。。

[balajiiitg]

I am unable to train CULANE small in condlanenet.py 'DataContainer' object has no attribute 'type'

[shuizaola]

你可以看下laneatt 他有python的量化代码，但是速度很慢。。。

能分享下culane的evaluate代码吗或者你的github也行？大佬！

[balajiiitg]

Hi I have seen the Laneatt but using CONDLANE There is problem in image_metas of condlanenet.py balaji

…

________________________________ From: shuizaola ***@***.***> Sent: 15 February 2022 12:31 To: aliyun/conditional-lane-detection ***@***.***> Cc: Balaji Rao K ***@***.***>; Comment ***@***.***> Subject: Re: [aliyun/conditional-lane-detection] Trained culane-small but could not get the performance described in the paper. (Issue #33) [CAUTION: This Email is from outside the Organization. Unless you trust the sender, Don’t click links or open attachments as it may be a Phishing email, which can steal your Information and compromise your Computer.] 你可以看下laneatt 他有python的量化代码，但是速度很慢。。。 能分享下culane的evaluate代码吗或者你的github也行？大佬！ ― Reply to this email directly, view it on GitHub<https://apc01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Faliyun%2Fconditional-lane-detection%2Fissues%2F33%23issuecomment-1039923934&data=04%7C01%7Cbalaji.raok%40hcl.com%7C8f640d7491bc43fe96d408d9f0510ffd%7C189de737c93a4f5a8b686f4ca9941912%7C0%7C0%7C637805053225268486%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=%2FvT455anzEHEzMLmrO8bcFJcJ%2FXtbRXHxEraQpEVjeQ%3D&reserved=0>, or unsubscribe<https://apc01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FAXS7TE3RH7WHZPOUPZC46ILU3H26LANCNFSM5I7AVIMQ&data=04%7C01%7Cbalaji.raok%40hcl.com%7C8f640d7491bc43fe96d408d9f0510ffd%7C189de737c93a4f5a8b686f4ca9941912%7C0%7C0%7C637805053225268486%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=Scw9AmXETD5yuNcUJKvCI8vhDW%2FgIHl0beHHgfWaTbo%3D&reserved=0>. Triage notifications on the go with GitHub Mobile for iOS<https://apc01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fapps.apple.com%2Fapp%2Fapple-store%2Fid1477376905%3Fct%3Dnotification-email%26mt%3D8%26pt%3D524675&data=04%7C01%7Cbalaji.raok%40hcl.com%7C8f640d7491bc43fe96d408d9f0510ffd%7C189de737c93a4f5a8b686f4ca9941912%7C0%7C0%7C637805053225268486%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=pt9UAXCf%2B1%2F4CoLHiCV3MyRhBBKvR0RCGDyIOnId3YM%3D&reserved=0> or Android<https://apc01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fplay.google.com%2Fstore%2Fapps%2Fdetails%3Fid%3Dcom.github.android%26referrer%3Dutm_campaign%253Dnotification-email%2526utm_medium%253Demail%2526utm_source%253Dgithub&data=04%7C01%7Cbalaji.raok%40hcl.com%7C8f640d7491bc43fe96d408d9f0510ffd%7C189de737c93a4f5a8b686f4ca9941912%7C0%7C0%7C637805053225268486%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=WLxYuwjtPTRoJ8LyDy5iiCAknHk4oK0HDgWxrGmuSVQ%3D&reserved=0>. You are receiving this because you commented.Message ID: ***@***.***> ::DISCLAIMER::

________________________________ The contents of this e-mail and any attachment(s) are confidential and intended for the named recipient(s) only. E-mail transmission is not guaranteed to be secure or error-free as information could be intercepted, corrupted, lost, destroyed, arrive late or incomplete, or may contain viruses in transmission. The e mail and its contents (with or without referred errors) shall therefore not attach any liability on the originator or HCL or its affiliates. Views or opinions, if any, presented in this email are solely those of the author and may not necessarily reflect the views or opinions of HCL or its affiliates. Any form of reproduction, dissemination, copying, disclosure, modification, distribution and / or publication of this message without the prior written consent of authorized representative of HCL is strictly prohibited. If you have received this email in error please delete it and notify the sender immediately. Before opening any email and/or attachments, please check them for viruses and other defects.

________________________________

[jyang68sh]

laneatt gets culane metrics by using inferred txt lane results.
@balajiiitg

[balajiiitg]

thanks by using
CUDA_VISIBLE_DEVICES=0,1,2,3 PORT=29001 tools/dist_train.sh configs/condlanenet/culane/culane_small_train.py 4
i am getting error dictionary object is not sussprictable
in function
def forward_train(self, img, img_metas, **kwargs):
gt_batch_masks = [m['gt_masks'] for m in img_metas]

[jyang68sh]

this error is straightforward.
Just debug img_meta

[balajiiitg]

ya i have debug that as image_metas.data.data[0][0]
output = self.backbone(img.type(torch.cuda.FloatTensor))? what is img.type has no attribute

[jyang68sh]

is img empty?

what is value of img?

[balajiiitg]

img has tensor with batch size of 8 but it has no attributes of img.device and img.type(torch.cuda.FloatTensor

…

________________________________ From: jyang68sh ***@***.***> Sent: 16 February 2022 11:50 To: aliyun/conditional-lane-detection ***@***.***> Cc: Balaji Rao K ***@***.***>; Mention ***@***.***> Subject: Re: [aliyun/conditional-lane-detection] Trained culane-small but could not get the performance described in the paper. (Issue #33) [CAUTION: This Email is from outside the Organization. Unless you trust the sender, Don’t click links or open attachments as it may be a Phishing email, which can steal your Information and compromise your Computer.] is img empty? — Reply to this email directly, view it on GitHub<https://apc01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Faliyun%2Fconditional-lane-detection%2Fissues%2F33%23issuecomment-1041148806&data=04%7C01%7Cbalaji.raok%40hcl.com%7C3accbea27ddc42f096ea08d9f1146c84%7C189de737c93a4f5a8b686f4ca9941912%7C0%7C0%7C637805892292262814%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=%2FeTNSFToqKOv0SeuNGLbvVmd7sFyAcFZdYLVTS9Y5H0%3D&reserved=0>, or unsubscribe<https://apc01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FAXS7TEY3ZBCLL66T4GM4ZT3U3M62RANCNFSM5I7AVIMQ&data=04%7C01%7Cbalaji.raok%40hcl.com%7C3accbea27ddc42f096ea08d9f1146c84%7C189de737c93a4f5a8b686f4ca9941912%7C0%7C0%7C637805892292419028%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=J%2B66WD1cJ0VCVJBrHuLGCvhGyGbdShDxmVKv4H1EGXM%3D&reserved=0>. Triage notifications on the go with GitHub Mobile for iOS<https://apc01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fapps.apple.com%2Fapp%2Fapple-store%2Fid1477376905%3Fct%3Dnotification-email%26mt%3D8%26pt%3D524675&data=04%7C01%7Cbalaji.raok%40hcl.com%7C3accbea27ddc42f096ea08d9f1146c84%7C189de737c93a4f5a8b686f4ca9941912%7C0%7C0%7C637805892292419028%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=yP5C9gdGXwicTgNvMV7x7ppBLiyDm%2Ft8li%2FqZuMb%2F7w%3D&reserved=0> or Android<https://apc01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fplay.google.com%2Fstore%2Fapps%2Fdetails%3Fid%3Dcom.github.android%26referrer%3Dutm_campaign%253Dnotification-email%2526utm_medium%253Demail%2526utm_source%253Dgithub&data=04%7C01%7Cbalaji.raok%40hcl.com%7C3accbea27ddc42f096ea08d9f1146c84%7C189de737c93a4f5a8b686f4ca9941912%7C0%7C0%7C637805892292419028%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=YJKK7ReYw0sG0vRfk9R0tyJHFybSlGFSwZlbjgoWTPo%3D&reserved=0>. You are receiving this because you were mentioned.Message ID: ***@***.***> ::DISCLAIMER::

________________________________ The contents of this e-mail and any attachment(s) are confidential and intended for the named recipient(s) only. E-mail transmission is not guaranteed to be secure or error-free as information could be intercepted, corrupted, lost, destroyed, arrive late or incomplete, or may contain viruses in transmission. The e mail and its contents (with or without referred errors) shall therefore not attach any liability on the originator or HCL or its affiliates. Views or opinions, if any, presented in this email are solely those of the author and may not necessarily reflect the views or opinions of HCL or its affiliates. Any form of reproduction, dissemination, copying, disclosure, modification, distribution and / or publication of this message without the prior written consent of authorized representative of HCL is strictly prohibited. If you have received this email in error please delete it and notify the sender immediately. Before opening any email and/or attachments, please check them for viruses and other defects.

________________________________

[jyang68sh]

if img has tensor value then it is not the reason for the error.

i am getting error dictionary object is not sussprictable

what is m in for m in img_meta?

[balajiiitg]

m consists on only gt_masks

…

________________________________ From: jyang68sh ***@***.***> Sent: 16 February 2022 11:56 To: aliyun/conditional-lane-detection ***@***.***> Cc: Balaji Rao K ***@***.***>; Mention ***@***.***> Subject: Re: [aliyun/conditional-lane-detection] Trained culane-small but could not get the performance described in the paper. (Issue #33) [CAUTION: This Email is from outside the Organization. Unless you trust the sender, Don’t click links or open attachments as it may be a Phishing email, which can steal your Information and compromise your Computer.] if img has tensor value then it is not the reason for the error. i am getting error dictionary object is not sussprictable what is m in for m in img_meta? — Reply to this email directly, view it on GitHub<https://apc01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Faliyun%2Fconditional-lane-detection%2Fissues%2F33%23issuecomment-1041151861&data=04%7C01%7Cbalaji.raok%40hcl.com%7C41db1bf6363b4baa819f08d9f1155399%7C189de737c93a4f5a8b686f4ca9941912%7C0%7C0%7C637805896170445776%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=BBpU8vKpCzyiFcEcbdenNmMA911ahs8U2WIn3J3PhRQ%3D&reserved=0>, or unsubscribe<https://apc01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FAXS7TE4AJQAQ4NWXAP6O2TTU3M7SZANCNFSM5I7AVIMQ&data=04%7C01%7Cbalaji.raok%40hcl.com%7C41db1bf6363b4baa819f08d9f1155399%7C189de737c93a4f5a8b686f4ca9941912%7C0%7C0%7C637805896170445776%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=O0I0aEWr1s9o%2Bu8VcCefLfo5GURjiqwUvsTpDIVb1Ms%3D&reserved=0>. Triage notifications on the go with GitHub Mobile for iOS<https://apc01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fapps.apple.com%2Fapp%2Fapple-store%2Fid1477376905%3Fct%3Dnotification-email%26mt%3D8%26pt%3D524675&data=04%7C01%7Cbalaji.raok%40hcl.com%7C41db1bf6363b4baa819f08d9f1155399%7C189de737c93a4f5a8b686f4ca9941912%7C0%7C0%7C637805896170445776%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=yfFl1vuObb%2BaySrtBmAMGYOzJs5adpsnVLAkC%2FqXank%3D&reserved=0> or Android<https://apc01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fplay.google.com%2Fstore%2Fapps%2Fdetails%3Fid%3Dcom.github.android%26referrer%3Dutm_campaign%253Dnotification-email%2526utm_medium%253Demail%2526utm_source%253Dgithub&data=04%7C01%7Cbalaji.raok%40hcl.com%7C41db1bf6363b4baa819f08d9f1155399%7C189de737c93a4f5a8b686f4ca9941912%7C0%7C0%7C637805896170445776%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=1NqcaQug9PX2uaR5n1zkDq5EJwgNu67cxuRDOcx1FTc%3D&reserved=0>. You are receiving this because you were mentioned.Message ID: ***@***.***> ::DISCLAIMER::

________________________________ The contents of this e-mail and any attachment(s) are confidential and intended for the named recipient(s) only. E-mail transmission is not guaranteed to be secure or error-free as information could be intercepted, corrupted, lost, destroyed, arrive late or incomplete, or may contain viruses in transmission. The e mail and its contents (with or without referred errors) shall therefore not attach any liability on the originator or HCL or its affiliates. Views or opinions, if any, presented in this email are solely those of the author and may not necessarily reflect the views or opinions of HCL or its affiliates. Any form of reproduction, dissemination, copying, disclosure, modification, distribution and / or publication of this message without the prior written consent of authorized representative of HCL is strictly prohibited. If you have received this email in error please delete it and notify the sender immediately. Before opening any email and/or attachments, please check them for viruses and other defects.

________________________________

[balajiiitg]

can we have quick teams call ?

…

________________________________ From: Balaji Rao K ***@***.***> Sent: 16 February 2022 12:10 To: aliyun/conditional-lane-detection ***@***.***>; aliyun/conditional-lane-detection ***@***.***> Cc: Mention ***@***.***> Subject: Re: [aliyun/conditional-lane-detection] Trained culane-small but could not get the performance described in the paper. (Issue #33) m consists on only gt_masks

________________________________ From: jyang68sh ***@***.***> Sent: 16 February 2022 11:56 To: aliyun/conditional-lane-detection ***@***.***> Cc: Balaji Rao K ***@***.***>; Mention ***@***.***> Subject: Re: [aliyun/conditional-lane-detection] Trained culane-small but could not get the performance described in the paper. (Issue #33) [CAUTION: This Email is from outside the Organization. Unless you trust the sender, Don’t click links or open attachments as it may be a Phishing email, which can steal your Information and compromise your Computer.] if img has tensor value then it is not the reason for the error. i am getting error dictionary object is not sussprictable what is m in for m in img_meta? — Reply to this email directly, view it on GitHub<https://apc01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Faliyun%2Fconditional-lane-detection%2Fissues%2F33%23issuecomment-1041151861&data=04%7C01%7Cbalaji.raok%40hcl.com%7C41db1bf6363b4baa819f08d9f1155399%7C189de737c93a4f5a8b686f4ca9941912%7C0%7C0%7C637805896170445776%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=BBpU8vKpCzyiFcEcbdenNmMA911ahs8U2WIn3J3PhRQ%3D&reserved=0>, or unsubscribe<https://apc01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FAXS7TE4AJQAQ4NWXAP6O2TTU3M7SZANCNFSM5I7AVIMQ&data=04%7C01%7Cbalaji.raok%40hcl.com%7C41db1bf6363b4baa819f08d9f1155399%7C189de737c93a4f5a8b686f4ca9941912%7C0%7C0%7C637805896170445776%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=O0I0aEWr1s9o%2Bu8VcCefLfo5GURjiqwUvsTpDIVb1Ms%3D&reserved=0>. Triage notifications on the go with GitHub Mobile for iOS<https://apc01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fapps.apple.com%2Fapp%2Fapple-store%2Fid1477376905%3Fct%3Dnotification-email%26mt%3D8%26pt%3D524675&data=04%7C01%7Cbalaji.raok%40hcl.com%7C41db1bf6363b4baa819f08d9f1155399%7C189de737c93a4f5a8b686f4ca9941912%7C0%7C0%7C637805896170445776%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=yfFl1vuObb%2BaySrtBmAMGYOzJs5adpsnVLAkC%2FqXank%3D&reserved=0> or Android<https://apc01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fplay.google.com%2Fstore%2Fapps%2Fdetails%3Fid%3Dcom.github.android%26referrer%3Dutm_campaign%253Dnotification-email%2526utm_medium%253Demail%2526utm_source%253Dgithub&data=04%7C01%7Cbalaji.raok%40hcl.com%7C41db1bf6363b4baa819f08d9f1155399%7C189de737c93a4f5a8b686f4ca9941912%7C0%7C0%7C637805896170445776%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=1NqcaQug9PX2uaR5n1zkDq5EJwgNu67cxuRDOcx1FTc%3D&reserved=0>. You are receiving this because you were mentioned.Message ID: ***@***.***> ::DISCLAIMER::

________________________________ The contents of this e-mail and any attachment(s) are confidential and intended for the named recipient(s) only. E-mail transmission is not guaranteed to be secure or error-free as information could be intercepted, corrupted, lost, destroyed, arrive late or incomplete, or may contain viruses in transmission. The e mail and its contents (with or without referred errors) shall therefore not attach any liability on the originator or HCL or its affiliates. Views or opinions, if any, presented in this email are solely those of the author and may not necessarily reflect the views or opinions of HCL or its affiliates. Any form of reproduction, dissemination, copying, disclosure, modification, distribution and / or publication of this message without the prior written consent of authorized representative of HCL is strictly prohibited. If you have received this email in error please delete it and notify the sender immediately. Before opening any email and/or attachments, please check them for viruses and other defects.

________________________________

[lovelydjj]

python tools/train.py configs/condlanenet/curvelanes/curvelanes_large_train.py
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [1, 256, 10, 25]], which is output 0 of ReluBackward0, is at version 1; expected version 0 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).

[haerrel]

I am unable to train CULANE small in condlanenet.py 'DataContainer' object has no attribute 'type'

I only get this error with dist_train.sh. I do not get the error when executing python tools/train.py configs/condlanenet/culane/culane_small_train.py. But with this command i get another error, which is the same as for @lovelydjj:

python tools/train.py configs/condlanenet/curvelanes/curvelanes_large_train.py
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [1, 256, 10, 25]], which is output 0 of ReluBackward0, is at version 1; expected version 0 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).

i was able to fix it with this #30