Open Lonicer opened 1 year ago
Can you reproduce the results for anet and Charades? @Lonicer
Can you reproduce the results for anet and Charades? @Lonicer
He did not give the anet data reading file. I only conducted experiments on the Charades data set, and it felt much lower, as follows: { "brief": { "MR-full-R1@0.5": 49.87, "MR-full-R1@0.7": 26.83, "MR-full-R5@0.5": 65.46, "MR-full-R5@0.7": 34.35, "MR-full-mAP": 28.54, "MR-full-mAP@0.5": 57.39, "MR-full-mAP@0.75": 24.63, "MR-middle-mAP": 31.58, "MR-short-mAP": 27.73 },
Can you also share the opt.json file for that result?
Here are my reproduced results for ANet which are much below their reported ones. For visual features, we used sub_activitynet_v1-3.c3d.hdf5. "MR-full-R1@0.5": 39.28, "MR-full-R1@0.7": 22.05, "MR-full-R5@0.5": 66.19, "MR-full-R5@0.7": 40.32, "MR-full-mAP": 27.64, "MR-full-mAP@0.5": 50.95, "MR-full-mAP@0.75": 24.84, "MR-middle-mAP": 24.05, "MR-short-mAP": 15.1
Are you using start_end_dataset.py provided by EaTR for data reading operations? Can hdf5 data be read? My Charades-STA options file is as follows. opt.zip
Yes right, but since self.feat_anet is not defined in the implementation, I defined with hd5 file by myself. And by the way, is 0.33333 right for the clip length? Isn't it should be 0.5 when using I3d?
Sorry for the late reply and confusion on the code. The data uploading code for the Charades-STA and ANet Caption datasets is the old version. I will update clear version of the code and config file as soon as possible.
Are you using start_end_dataset.py provided by EaTR for data reading operations? Can hdf5 data be read? My Charades-STA options file is as follows. opt.zip
Hello! Did you successfully run the code on the charades dataset? Could you tell me your wechat or email address? I want to ask you something about the code running!
Sorry for bothering, I wonder why the ActivityNet Caption Link can not be donwloaded and where are the txt_feature?
Sorry for the late reply and confusion on the code. The data uploading code for the Charades-STA and ANet Caption datasets is the old version. I will update clear version of the code and config file as soon as possible.
Sorry for bothering, I wonder if you could give the config file of charades-STA?
Your work is very good, I want to know more about it, but your paper doesn’t seem to show how Anet’s CLIP text feature is obtained, and Anet’s visual feature is: sub_activitynet_v1-3.c3d.hdf5?