danqu130 / dceiflow

Learning Dense and Continuous Optical Flow from an Event Camera (TIP 2022)

dceiflow's Issues

Can you also share the event simulation code to me, please?

Can you also share the event simulation code to me ([email protected]), please?

Code about visualizing flow

Thanks for your release.
But how can I get the flow images like those in your paper? I did't find the code to visualize the flow. I've implemented the code by myself, but I can't get the same results. It would be very helpful if you can send me the code. My email is [email protected]. Thanks a lot.

Test on indoor_flying3 fails

test on indoor_flying3 fails and here is the error information.

-- Process 0 terminated with the following error:
Traceback (most recent call last):
File "/home/mmspg/anaconda3/lib/python3.10/site-packages/torch/multiprocessing/spawn.py", line 69, in _wrap
fn(i, *args)
File "/home/mmspg/Desktop/DCEIFlow/main.py", line 229, in test
scores = evaluates(args, model, test_sets, test_setnames, metric_fun, logger=logger)
File "/home/mmspg/Desktop/DCEIFlow/evaluate.py", line 54, in evaluates
metric = evaluate(args, model, val_loader, name, metric_fun, logger=logger)
File "/home/mmspg/Desktop/DCEIFlow/evaluate.py", line 80, in evaluate
for index, batch in enumerate(dataloader):
File "/home/mmspg/anaconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 634, in next
data = self._next_data()
File "/home/mmspg/anaconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 1346, in _next_data
return self._process_data(data)
File "/home/mmspg/anaconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 1372, in _process_data
data.reraise()
File "/home/mmspg/anaconda3/lib/python3.10/site-packages/torch/_utils.py", line 644, in reraise
raise exception
AssertionError: Caught AssertionError in DataLoader worker process 0.
Original Traceback (most recent call last):
File "/home/mmspg/anaconda3/lib/python3.10/site-packages/torch/utils/data/_utils/worker.py", line 308, in _worker_loop
data = fetcher.fetch(index)
File "/home/mmspg/anaconda3/lib/python3.10/site-packages/torch/utils/data/_utils/fetch.py", line 51, in fetch
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/home/mmspg/anaconda3/lib/python3.10/site-packages/torch/utils/data/_utils/fetch.py", line 51, in
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/home/mmspg/Desktop/DCEIFlow/utils/datasets/MVSEC.py", line 182, in getitem
final_flow = generate_corresponding_gt_flow(flows, flows_ts, image1_ts, next_ts)
File "/home/mmspg/Desktop/DCEIFlow/utils/datasets/MVSEC_utils.py", line 102, in generate_corresponding_gt_flow
assert flow_length == len(flows_ts) - 1
AssertionError

Can someone help to take a look at this? Thank you,

Code about simulating events on FlyingChairs2

Nice work!
Can you release the codes about simulating events on FlyingChairs2 or release the event datasets for easy use.

Training on the MVSEC dataset

Your work is very meaningful!
After reading paper and codes, I am confused about the contents in Table IV. Table IV show the results of two baselines and DCEIFlow on the MVSEC dataset. My confusion is how you trained models on MVSEC:

1. How the training sequence was divided? The 'valid_time_index ' of outdoor_day2 is set to be [4375,7002]. Should I only use them to train?
2. Which is the Epoch to be set? 200 or 300? And how the other hyperparameters configured?
3. Can you release the weights of DCEIFlow trained on MVSEC?

I am looking forward to your reply.

Outcomes of E-RAFT in Table II

Hello!
I have some questions about the outcomes of E-RAFT trained on DSEC in Table II:

1. Which weight is used? The one published by E-RAFT or the one trained by yourself?
2. How is the MVSEC data processed? Is it the same as E-RAFT, or did you make some changes?
3. Did you use the warm start mentioned in ERAFT during testing?

I am looking forward to your reply! And I would greatly appreciate it if you could share your testing code on ERAFT.