uttaranb127 / step Goto Github PK

Hi,

Thank you very much for sharing your code.

I am wondering why you have 6 labels in the dataset but in the paper 4 emotions are used.

Thanks!
Xingchen

Dear Uttaran, I want to understand, how have you created a neighbor_link (in graph.py sctript) for you dataset.
Can you explain me your idea on openpose example please?

num_node = 18
self_link = [(i, i) for i in range(self.num_node)]
neighbor_link = [(4, 3), (3, 2), (7, 6), (6, 5), (13, 12), (12, 11),
                 (10, 9), (9, 8), (11, 5), (8, 2), (5, 1), (2, 1),
                 (0, 1), (15, 0), (14, 0), (17, 15), (16, 14)]
edge = self_link + neighbor_link
center = 1

Hi! In the paper you name 3 models, with STEP+Aug being the one with the best accuracy. Is STEP+Aug equivalent with the model on the classifier_stgcn_real_and_synth folder? Also, in the mainf file for classifier_stgcn_real_and_synth, the synth filename is 'CVAEGCN_best' but I couldn't find it in the provided dataset nor inside the repository. I appreciate your help!

Hello,

I read this paper recently and wanted to try my hands on it. Thank you for providing the code for the same.

How to obtain the affective features of a custom dataset?

Regards
Abhishek

Hello!
Thanks for your working very much!!!

I'm sorry this questions can be annoying.

I wanted to cite your work in my project and meet some problems. Since I have no previous experience with machine learning and python, I don't know how to use this code to predict the corresponding emotion based on a video of a person walking.
Could you please tell me:

1. Whether these codes need to be run based on extracted gait data instead of gait video.

2. If I want to input a gait video to get his emotions, such as happy and angry. What more needs to be done.

Hi, Uttaran
I have some question for your paper and code.

1. In your paper, training set:validation set:test set=7:2:1
   But actually, in the open code, it is Training:Test = 9:1. which one should I use?

2. Your paper describes that the addition of 29-dim affective features will give a great improvement in the classification accuracy, but when I run your code, the experimental results are instead reduced with a great overfitting, what is the reason for this?

Thank you! : )

Hi,

Thank you for your work and code!

I have a question on Fig. 6 of your paper.

In the caption, you said you were showing the results over the 3177 gaits. But in the figure, the total number of 4 classes (Angry, 1073, Happy 565, Sad 65 and Neutral 232) are 1935 not 3177. In adition, I think you trained the model on E-gaits dataset. In this case, are you reporting results on all the E-gaits not only on the test set? Because I think your test set does not have so many samples.

Finally, as shown in Fig.6, your sample classes are highly imbalanced. I am wondering if your training set is also imbalanced, and in this case, how much sense the accuracy can be used as the evaluation metric.

Sorry for so many questions. I am not sure if I misunderstood something. Could you please kindly help to clarify?

Many thanks!
Xingchen

Hello,

Thank you for your incredible work. I have a question regarding the data provided for training (code).

1. From the paper I understand that for emotion classification, the gait + affective features are used for classifying? But from the code it seems that just the affective features are split into data_training and data_testing or am I mistaken?

Thanks

Hello,Uttaran!
I was wondering if the full dataset of 4227 samples is available now?
If not, when will it be announced?
Im looking forward to your reply

Hello sir, I have found that the scale of dataset depicted in the paper is different from the one in the website.
In the paper, there are 5227 = 4227(real) + 1000(synthetic). However, I get 6177=2177(real) + 4000(stnthetic) in the website. Please tell me the reason.

Hi,

Thanks for the code and dataset you have offered! I find myself really interested in your work!

I have some questions about the dataset as follows:

1. I want to visualize the motion sequences in the dataset, but I find that the root joint (joint #0) is not at the origin of the coordinates, which is different from the ELMD dataset setting. (visualization in the first figure below) Also, when I try relocating the root joint to the origin, the scales and visual angles of the gaits vary a lot. (visualization in the second figure below) I wonder whether I need to do some preliminary works before visualizing the gaits.

2. The synthetic gaits in the dataset seem a little strange. After performing visualization, it looks like the person is dragging his/her legs rather than walking.

3. I wonder how the emotion labels are annotated. Are they annotated by pretrained models, or annotated manually?

Many thanks and best wishes!
Shuhong

Hello Uttaran,i have some questions about the datasets that you posted on GoogleCloud.
i found that the real gaits on GoogleCloud have 112+1048 angry samples、33+454 neutral samples、78+254 happy samples and 119+79 sad samples respectively from 342 were collected by you and the remaining 1,835 were taken from the ELMD.
Firstly,In the paper, there are 4227 real samples. However,the number of neutral/angry samples became less.Is there much difference between the datasets in parper and that of GoogleCloud?
Then,the input gaits were converted to 21-joint skeletons in the paper.but that of github code is 16.
Finally,may i ask when you can upload the latest datasets.
Looking forward to your reply

There a lot of problems when I run the code. Can you provide information about your experimental environment? Thanks!

Dear Uttaran, I want to use a 2D pose dataset instaed of a 3D, that you use in this work. Tell me please what does 'in_channels' there https://github.com/UttaranB127/STEP/blob/master/classifier_stgcn_real_only/net/classifier.py#L19 mean?

Is it responsible for 3 (RGB) image channels or it is about 3 dimensional input data and if I want to use a 2D dataset, I have to set in_channels=2?