In large systems/complicated situations, I prefer using Quantum Espresso, at least to generate the intermediate images. You only need to start the calculations in it, and then export the obtained images' XYZ to whatever format you want.

You only need to provide initial and the final state  (confirming one to mapping of atoms in the two state) of the system. Intermediate images are generated by QE by linear interpolation as a starting guess.

Also, use more number of images. 1 is too scanty I would say. User something like 5, 7, 9 etc images.

It works fine for me. See if you can do that.

A single image may indeed be too few for a normal NEB calculation, as the image might end up not very close to the maximum. However, using many images as  Vaibhav suggested will make ionic convergence very slow. A solution may lie in the use of the climbing image NEB method, which always puts one image (even it it is the only one) at the top of the barrier. There is plug-in code for CI-NEB available for VASP, see

http://theory.cm.utexas.edu/vtsttools/

I guess

   nebavoid.pl

is the script that that Bharat is looking for.

Very handy and useful, Peter. Thanks.

http://theory.cm.utexas.edu/vtsttools/scripts.html

Thanks guys for the replies. 

@Vaibhav, thanks for the link.  I will give the script a try. I initially started with 4 or 5 images, but again I faced the convergence issue with not even a single electronic convergence. Also my supercells are big (>250 atoms due to interface), hence I was spending so much time playing with VASP parameters to get convergence. So I went down to two and one images because I didn't understand the source of trouble initiially.

@Peter, Yes, I read about CI-NEB method and have it already installed. I was using it initially, but now  planning to use it once I have convergence with NEB method. 

By the way, the method I suggested as option 2 is working. I generated 16 images and discarded all except the 2 images that are close to the end points. The calculations haven't finished yet, but I am not noticing major problems with convergence.  

Another question I have is on how to refine the NEB calculations.  I see two ways to proceed. Once the images converge to the minimum energy path (MEP), I can generate new images by interpolating the optimized images again. 1) After generating the new images, shall I keep the prior images converged in earlier run as end points and converge the new images as suggested in vasp manual? 2) Or do I incorporte the old and new images into a single NEB calculation with tthe original end points? Do the converged images move on MEP due to the inclusion of additional images? I understand option (1) is easy on resources and also may be efficient because I don't have to waste cpu time on already optimized images.

Thanks again.