Closed ily-R closed 4 years ago
Hello @ily-R , thanks for asking. I think you are right, we need to change this to modulatedatt. I think it is a bug when we were renaming modules for code publication. We will update the code asao
Hello @ily-R , the codes are updated, and we have newly implemented results and weights for Places-LT as well. Please check it out. Thanks.
Thank you ! I will check it out.
On Thu, Mar 5, 2020, 08:19 Zhongqi Miao notifications@github.com wrote:
Hello @ily-R https://github.com/ily-R , the codes are updated, and we have newly implemented results and weights for Places-LT as well. Please check it out. Thanks.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/zhmiao/OpenLongTailRecognition-OLTR/issues/53?email_source=notifications&email_token=AFNM2UTYHCO4ALVITX4KY4LRF5G6PA5CNFSM4LAOPFY2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEN4AAAQ#issuecomment-595066882, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFNM2UQ2YRLXNXW34IF4YMDRF5G6PANCNFSM4LAOPFYQ .
https://github.com/zhmiao/OpenLongTailRecognition-OLTR/blob/4dc25ea546bbcb0f84033f8187509a9d0b96fab3/run_networks.py#L76
Seeing this line, Stage1 config and Stage2 config, I noticed that embedded gaussian non-local filtering is not trained at all as "selfatt" is not present in ModulatedAttLayer. In Stage 1 we don't use the modulated attention https://github.com/zhmiao/OpenLongTailRecognition-OLTR/blob/4dc25ea546bbcb0f84033f8187509a9d0b96fab3/config/Places_LT/stage_1.py#L19
In Stage 2 we use it: https://github.com/zhmiao/OpenLongTailRecognition-OLTR/blob/4dc25ea546bbcb0f84033f8187509a9d0b96fab3/config/Places_LT/stage_2_meta_embedding.py#L20 initialize it, but then fix all the convolution layers weights due to: https://github.com/zhmiao/OpenLongTailRecognition-OLTR/blob/4dc25ea546bbcb0f84033f8187509a9d0b96fab3/config/Places_LT/stage_2_meta_embedding.py#L26
and the above if condition.
Should we change "selfatt" to "modulatedatt"