jaywongwang / i3d-feature-extractor Goto Github PK

Hello JaywongWang,

Nice job, it's quite amazing and helpful, yet I have ran into a problem of 'mismatch between checkpoint files and model'.

I downloaded checkpoint files from "https://github.com/deepmind/kinetics-i3d/tree/master/data/checkpoints/rgb_imagenet" as you mentioned, but when I executed "saver.restore(sess, _CHECKPOINT_PATHS['rgb_imagenet'])" from 「feature_extractor_frm.py」 it says :「Key Conv3d_1a_7x7/batch_norm/beta not found in checkpoint
[[node save/RestoreV2 (defined at :10) ]]」.

Could you please lend me a hand? Thank you very much!

Best,
Tamio

Hello I want to know ur environment. can you inform ur versions info?(python,tensorflow, numpy, sonet, PIL)

Hi!If n_frames represents the frames of a video, How can I extract features of size n_frames*1024 with rgb_videos as the input?

when i restore the rgb_imagenet model, the layer name maybe wrong, I change into

for variable in tf.global_variables():
rgb_variable_map['RGB/'+variable.name.replace(':0', '')] = variable

then the name is right;

but I meet a new question:

InvalidArgumentError (see above for traceback): Restoring from checkpoint failed. This is most likely due to a mismatch between the current graph and the graph from the checkpoint. Please ensure that you have not altered the graph expected based on the checkpoint. Original error:

Assign requires shapes of both tensors to match. lhs shape= [1,1,1,1,16] rhs shape= [16]
[[node save/Assign_75 (defined at /data1/wlx/project/I3D-Feature-Extractor-master/feature_extractor_frm.py:41) ]]

how can i solve this problem?
thanks

When I run the program, I run into the contents of the feature_Extractor () function.
Always throw the following error:
Process finished with exit code -1073741819 (0xC0000005)
What is the reason
Also, can you help me mark the content of each input file (OUTPUT_FEAT_DIR, VIDEO_PATH_FILE, VIDEO_DIR)?